A spatio-temporal data compression method and system based on lightweight processing

Abstract

The application discloses a kind of spatiotemporal data compression method and system based on light processing, it is related to image data processing technical field, the method includes: based on the spatiotemporal data to be compressed, establishes standardization data entry, utilizes time axis segmentation to form window set, according to frequency energy distribution, window is periodically discriminated, constructs spatiotemporal block, and based on global error budget, error control parameter table is established;Generate time coding stream, integer is carried out to three-dimensional coordinate and attribute value, and space coding stream is generated through double difference and run-length encoding, grid data is folded and lightened with edge, and texture compression is generated Joint encoding result;Based on joint encoding result, compressed data is divided into base layer and several enhancement layers, establish the relevance layered storage structure and block level index table, to generate the compressed data of streamable transmission.The application can utilize the internal correlation of spatiotemporal data to carry out collaborative compression.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and more specifically, to a spatiotemporal data compression method and system based on lightweight processing. Background Technology

[0002] With the rapid development of 3D image processing technology, dynamic 3D models containing spatiotemporal information are widely used in smart cities, land resource management, virtual reality, and other fields. Taking real-scene 3D geographic models as an example, each region has different surface conditions at different times, and each period of mapping data contains complete 3D geometric information and texture data. Continuous updates create a dynamic effect of surface changes. However, a single region's real-scene 3D model file often contains massive amounts of 3D data, with a size reaching hundreds of megabytes or even several gigabytes. Faced with the limitations of computing resources and network bandwidth in geographic information systems, how to efficiently compress this type of 3D image data containing spatiotemporal changes, while ensuring visual quality and achieving fast transmission and real-time rendering, has become a significant challenge in the field of image data processing technology.

[0003] Existing 3D image data compression techniques mainly fall into two categories: geometric compression and texture compression. Geometric compression techniques reduce the number of vertices and faces through mesh simplification algorithms; common methods include quadratic error metric simplification and progressive mesh coding. Texture compression techniques encode texture data using standard image compression formats such as JPEG and DXT. Furthermore, some research has proposed 3D model compression methods based on wavelet transform and geometric data compression algorithms based on predictive coding, achieving certain results in optimizing the storage space of static 3D image models.

[0004] However, existing image data compression techniques have significant shortcomings when processing dynamic 3D data containing spatiotemporal information. First, existing methods typically process time and spatial information separately, lacking comprehensive utilization of the inherent correlation between spatiotemporal data and failing to fully exploit redundant information between consecutive frames for collaborative compression, resulting in limited overall compression efficiency. Second, traditional compression algorithms often employ fixed processing strategies, which significantly degrade compression performance when faced with dynamic spatiotemporal data such as complex terrain changes, making it difficult to meet the requirements of real-time processing and efficient transmission.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] To address the problems in related technologies, this invention proposes a spatiotemporal data compression method and system based on lightweight processing. It has the advantages of collaborative compression by utilizing the inherent correlation of spatiotemporal data, supporting layered encapsulation and streaming transmission, thereby solving the problems of limited compression efficiency and difficulty in balancing compression ratio and visual quality in existing technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a spatiotemporal data compression method based on lightweight processing is provided, the spatiotemporal data compression method based on lightweight processing includes:

[0009] S1. Based on the spatiotemporal data to be compressed, standardized data entries are established through preprocessing, window sets are formed by segmenting the time axis, the windows are periodically judged according to the frequency domain energy distribution, spatiotemporal blocks are constructed, and an error control parameter table is established based on the global error budget.

[0010] S2. Generate a time-coded stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial-coded stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results.

[0011] S3. Based on the joint encoding results, the compressed data is divided into a basic layer and several enhancement layers. An associative hierarchical storage structure and a block-level index table are established to generate compressed data that can be streamed.

[0012] Furthermore, the spatiotemporal data to be compressed includes at least one of vertex data, coordinate data, normal data, color data, texture data, terrain change parameter data, and trajectory point data. The preprocessing to establish standardized data entries includes: identifying the data type of the spatiotemporal data to be compressed and determining the included data elements; uniformly converting the data elements to the same coordinate system and time axis, removing outliers exceeding a preset threshold range to obtain initial data; uniformly processing the sampling rate of the data sequence in the initial data and aligning various attributes to obtain aligned data; and organizing the aligned data into standardized data entries, which include time points, 3D coordinates, attribute values, and grid data.

[0013] Furthermore, forming a window set by segmenting the time axis includes: calculating the rate of change of the three-dimensional coordinates in the standardized data entries between adjacent time points to represent the magnitude of spatial coordinate change per unit time; setting a dynamic window length based on the rate of change, using a window shorter than the preset length for regions with a rate of change higher than a first preset threshold, and using a window longer than the preset length for regions with a rate of change lower than a second preset threshold; segmenting the time axis using a sliding window with a dynamic window length to generate a window set; and establishing a window-segment index for each window in the window set to facilitate block-based processing of standardized data entries.

[0014] Furthermore, the periodicity of windows is determined based on the frequency domain energy distribution, and the spatiotemporal block is constructed by: performing Fourier transform on the time series data of standardized data entries within each window in the window set to calculate the frequency domain energy distribution; determining the energy proportion of the target frequency component in the frequency domain energy distribution based on the spectral threshold, marking the window as a periodic segment when the proportion exceeds the spectral threshold, otherwise marking it as a non-periodic segment, thus obtaining the window label; determining the spatial neighborhood range within each window of the window set, where the spatial neighborhood range is the clustered region of three-dimensional coordinates in adjacent standardized data entries; and combining the window, the spatial neighborhood range, and the window label to form a spatiotemporal block as the basic unit for compression processing.

[0015] Furthermore, establishing an error control parameter table based on the global error budget includes: setting a global error budget to represent the maximum allowable error after compressed reconstruction; allocating the global error budget into a spatial error budget and a temporal error budget according to the window label and spatial neighborhood range; determining the coordinate integerization step size based on the spatial error budget; determining the temporal quantization threshold based on the temporal error budget; detecting the rate of change of three-dimensional coordinates and attribute values ​​in standardized data entries, marking time points where the rate of change exceeds a preset mutation threshold as anchor points, and generating an anchor point set; and combining the coordinate integerization step size, the temporal quantization threshold, the global error budget, and the anchor point set to form the error control parameter table.

[0016] Furthermore, generating a time-coded stream based on the spatiotemporal block and error control parameter table includes: determining the type of the current spatiotemporal block based on the window label; for spatiotemporal blocks marked as periodic segments, extracting the amplitude and phase parameters of the target frequency components through frequency domain transformation to form a frequency domain parameter set; for spatiotemporal blocks marked as non-periodic segments, constructing a time series prediction model to calculate the predicted value and comparing it with the actual value to obtain the residual sequence; performing dead-zone quantization on the frequency domain parameter set and the residual sequence according to the time quantization threshold to control quantization error; retaining the original data at the corresponding time point based on the anchor point set without quantization processing; and organizing the quantized frequency domain parameter set and the residual sequence into a time-coded stream.

[0017] Furthermore, the process of integerizing standardized data entries and generating a spatial encoded stream through double difference and run-length encoding includes: quantizing the three-dimensional coordinates and attribute values ​​of standardized data entries according to the coordinate integerization step size to obtain an integer sequence; performing first-order difference calculation on the integer sequence to obtain a first-order difference sequence, and performing second-order difference calculation to obtain a second-order difference sequence; analyzing the sparsity characteristics of the second-order difference sequence to determine continuous repeating values; encoding the continuous repeating values ​​using run-length encoding to obtain a run-length encoded sequence; and further compressing the run-length encoded sequence by entropy encoding to generate a spatial encoded stream.

[0018] Furthermore, edge-folding lightweighting and texture compression are performed on the grid data in the standardized data entries to generate joint encoding results. This includes: calculating the quadratic error metric for each edge of the grid data, selecting edges for folding operations according to the order of error metric values ​​to obtain a simplified grid; using the spatial error budget as a distance constraint parameter, calculating the maximum distance between the current simplified grid and the original grid during the folding process, stopping folding when the maximum distance reaches the constraint parameter, and obtaining simplified grid structure data; constructing a two-dimensional mapping for the texture information of the grid data, calculating the pixel occupancy rate of each region, merging texture regions with occupancy rates below a preset threshold into continuous regions to form an optimized texture atlas; applying a block compression algorithm to encode the optimized texture atlas to generate compressed texture data; organizing the simplified grid structure data and compressed texture data into a geometric texture lightweight stream; and combining the temporal encoding stream, spatial encoding stream, and geometric texture lightweight stream into a joint encoding result.

[0019] Furthermore, based on the joint coding results, the compressed data is divided into a base layer and several enhancement layers. An associative hierarchical storage structure and a block-level index table are established to generate streamable compressed data. This includes: extracting data from the temporal coded stream, spatial coded stream, and lightweight geometric texture stream that satisfy the spatial error budget and temporal error budget to form the base layer; organizing the unextracted data from the temporal coded stream, spatial coded stream, and lightweight geometric texture stream into several enhancement layers; establishing a temporal-space associative hierarchical storage structure based on the window, spatial neighborhood range, and window label of the spatiotemporal block to record the correspondence between each layer and the spatiotemporal block; establishing a block-level index table based on the associative hierarchical storage structure to record the index information of the window, spatial neighborhood range, and window label corresponding to each layer; storing simplified mesh structure data only once in the base layer for recurring mesh data, and storing reference identifiers pointing to the base layer in the enhancement layers; and encapsulating the base layer and enhancement layers using the block-level index table to generate streamable compressed data.

[0020] According to another aspect of the present invention, a spatiotemporal data compression system based on lightweight processing is also provided, the spatiotemporal data compression system based on lightweight processing comprising:

[0021] The preprocessing module is used to establish standardized data entries based on the spatiotemporal data to be compressed, form a window set by segmenting the time axis, periodically distinguish the windows according to the frequency domain energy distribution, construct spatiotemporal blocks, and establish an error control parameter table based on the global error budget.

[0022] The encoding generation module is used to generate a temporal encoding stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial encoding stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results.

[0023] The layered encapsulation module is used to divide compressed data into a base layer and several enhancement layers based on the joint encoding results, and to establish an associative layered storage structure and block-level index table to generate compressed data that can be streamed.

[0024] The beneficial effects of this invention are as follows:

[0025] (1) By establishing standardized data entries and an adaptive window mechanism based on the rate of change, this invention performs unified preprocessing and coupled modeling of spatiotemporal data, realizing the joint processing of time and space information, and effectively mining redundant information between consecutive frames. Compared with the traditional method of processing time and space information separately, this invention can make full use of the inherent correlation of spatiotemporal data for collaborative compression, significantly improving the overall compression efficiency. In particular, for dynamic models containing a large amount of three-dimensional data, such as real-scene three-dimensional geographic models, the compression rate is improved by more than 30%, solving the problem of low efficiency of existing compression technology when processing dynamic three-dimensional data containing spatiotemporal information.

[0026] (2) This invention can automatically identify the changing characteristics of spatiotemporal data and adopt different coding strategies through the periodic discrimination mechanism of frequency domain energy distribution and the error control parameter table. It adopts the frequency domain parameterization method for periodic segments and the prediction-residual method for non-periodic segments. At the same time, by combining various technical means such as coordinate integerization, double difference and edge folding lightweight, this invention forms an adaptive processing framework for different types of spatiotemporal data. It effectively overcomes the technical defects of traditional compression algorithms that use fixed processing strategies, which lead to a decrease in compression performance when facing dynamic spatiotemporal data with complex terrain and landform changes. The compressed data can be reduced to 15%-25% of the original data while ensuring visual quality, which greatly reduces the demand for geographic information system storage space and network bandwidth.

[0027] (3) The present invention designs a layered storage structure and block-level index table that combines the base layer and the enhancement layer, realizing the layered encapsulation and streaming transmission of compressed data. This allows the decoding end to load the corresponding layer of data as needed according to actual requirements and device performance. On low-performance devices such as thin clients, only the base layer data can be loaded to achieve fast rendering. On high-performance devices such as fat clients, the enhancement layer data can be loaded gradually to improve visual quality. At the same time, the anchor point mechanism is used to specially process the sudden change points to ensure the reconstruction accuracy of frames at important moments. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart illustrating a spatiotemporal data compression method based on lightweight processing according to an embodiment of the present invention;

[0030] Figure 2 This is a detailed implementation diagram of constructing a spatiotemporal block and an error control parameter table in a spatiotemporal data compression method based on lightweight processing according to an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of a spatiotemporal data compression system based on lightweight processing according to an embodiment of the present invention. Detailed Implementation

[0032] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0033] According to an embodiment of the present invention, a spatiotemporal data compression method and system based on lightweight processing are provided.

[0034] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a spatiotemporal data compression method based on lightweight processing is provided, the spatiotemporal data compression method based on lightweight processing includes:

[0035] S1. Based on the spatiotemporal data to be compressed, standardized data entries are established through preprocessing, window sets are formed by segmenting the time axis, the windows are periodically judged according to the frequency domain energy distribution, spatiotemporal blocks are constructed, and an error control parameter table is established based on the global error budget.

[0036] S2. Generate a time-coded stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial-coded stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results.

[0037] S3. Based on the joint encoding results, the compressed data is divided into a basic layer and several enhancement layers. An associative hierarchical storage structure and a block-level index table are established to generate compressed data that can be streamed.

[0038] In one embodiment, the spatiotemporal data to be compressed includes at least one of vertex data, coordinate data, normal data, color data, texture data, terrain variation parameter data, and trajectory point data; such as Figure 2 As shown, the preprocessing to establish standardized data entries includes: identifying the data type of the spatiotemporal data to be compressed and determining the included data elements; uniformly transforming the data elements to the same coordinate system and time axis, removing outliers exceeding a preset threshold range to obtain initial data; uniformly processing the sampling rate of the data sequence in the initial data and aligning various attributes to obtain aligned data; and organizing the aligned data into standardized data entries, which include time points, three-dimensional coordinates, attribute values, and grid data.

[0039] Specifically, in this invention, data type identification of the spatiotemporal data to be compressed is achieved by parsing the format identifier in the header of the data file. Taking common 3D model file formats in geographic information systems as examples, such as the binary structure in OSGB (.osgb) files and the texture link information in OBJ (.obj) formats, the data elements such as terrain feature points, vertices, and textures contained in the model can be directly identified. For XML-based formats such as GML (.gml) or KML (.kml), identification is achieved by parsing their XML tag structure. The unified transformation of coordinate systems uses the right-handed coordinate system as the standard reference system, and the data under different coordinate systems is uniformly transformed through matrix transformation. The time axis is uniformly measured in milliseconds; outlier removal is based on statistical methods, calculating the mean μ and standard deviation σ of each component of the 3D coordinates, and marking data points that deviate from the mean by more than 3σ as outliers and removing them; the sampling rate is uniformly processed for mapping data collected at different frame rates, using a cubic spline interpolation algorithm to resample the data on the time axis to a uniform 30 frames / second; attribute alignment is achieved by mapping attributes such as vertices, normals, and colors one-to-one according to their indices; the final standardized data entries are in the form of six-tuples <t,x,y,z,a,m>, where t represents the time point (milliseconds), x, y, and z represent the 3D coordinates (meters), a represents the vector of attribute values ​​such as color and normals, and m represents the topological structure of the mesh connection relationship.

[0040] In one embodiment, forming a window set by segmenting the time axis includes: calculating the rate of change of the three-dimensional coordinates in the standardized data entry between adjacent time points to represent the magnitude of change of spatial coordinates per unit time; setting a dynamic window length according to the rate of change, using a window shorter than the preset length for regions with a rate of change higher than a first preset threshold, and using a window longer than the preset length for regions with a rate of change lower than a second preset threshold; segmenting the time axis using a sliding window with a dynamic window length to generate a window set; and establishing a window-segment index for each window in the window set to facilitate block-based processing of standardized data entries.

[0041] Specifically, the rate of change of the three-dimensional coordinates in the standardized data entries between adjacent time points is calculated using the Euclidean distance metric. For time points t and t+1, the rate of change r(t) = ||p(t+1) - p(t)|| / Δt, where p(t) represents the three-dimensional coordinates at time point t, and Δt is the interval between adjacent time points. The dynamic window length is set based on the statistical distribution of the rate of change, and the average rate of change r for the entire sequence is calculated. avg and standard deviation r σ The first preset threshold is set to r. avg +r σ The second preset threshold is set to r. avg -r σ For regions with a rate of change higher than the first preset threshold, the window length is set to half the base window length (30 frames); for regions with a rate of change lower than the second preset threshold, the window length is set to twice the base window length. When the sliding window is segmented on the time axis, a 25% overlap is allowed between adjacent windows to ensure data continuity. The window-segment index uses a tuple <w id ,[t start ,t end The form is ]〉, where w id t is a unique identifier for the window. start and t end These represent the start and end times of the window, respectively.

[0042] In one embodiment, the periodicity determination of windows based on frequency domain energy distribution and the construction of spatiotemporal blocks include: performing Fourier transform on the time series data of standardized data entries within each window in the window set to calculate the frequency domain energy distribution; determining the energy proportion of the target frequency component in the frequency domain energy distribution based on a spectral threshold, marking the window as a periodic segment when the proportion exceeds the spectral threshold, otherwise marking it as a non-periodic segment, thus obtaining a window label; determining the spatial neighborhood range within each window of the window set, where the spatial neighborhood range is the clustered region of three-dimensional coordinates in adjacent standardized data entries; and combining the window, the spatial neighborhood range, and the window label to form a spatiotemporal block as the basic unit for compression processing.

[0043] Specifically, when performing Fourier transform on the time-series data of standardized data entries within each window in the window set, the x, y, and z coordinates of each vertex within the window are first treated as one-dimensional signals that vary with time. The Fast Fourier Transform (FFT) algorithm is applied to obtain the spectrum. Then, the spectral energy of each coordinate component is summed to obtain the comprehensive frequency domain energy distribution. The spectrum threshold is set to 70% of the total energy; that is, when the proportion of the maximum energy frequency component to the total energy exceeds 70%, it is determined to have obvious periodicity and marked as a periodic segment. The spatial neighborhood is determined using a spatial grid division method, dividing the three-dimensional space into cubic grids of 0.1 meters in size. Data points located in the same or adjacent grids are considered to be in the same spatial neighborhood. A breadth-first search algorithm is used to cluster the grids to obtain the clustered regions of the spatial neighborhood. The spatiotemporal block is ultimately defined by tuples <w id ,{s id The form},tag〉 represents the tag, where w id For window identifier, {s id} represents the set of spatial neighborhood identifiers, and tag is a periodic label (0 represents a non-periodic segment, and 1 represents a periodic segment).

[0044] In one embodiment, establishing an error control parameter table based on a global error budget includes: setting a global error budget to represent the maximum allowable error after compressed reconstruction; allocating the global error budget into a spatial error budget and a temporal error budget according to the window label and spatial neighborhood range; determining the coordinate integerization step size based on the spatial error budget; determining the temporal quantization threshold based on the temporal error budget; detecting the rate of change of three-dimensional coordinates and attribute values ​​in standardized data entries, marking time points where the rate of change exceeds a preset mutation threshold as anchor points, and generating an anchor point set; and combining the coordinate integerization step size, the temporal quantization threshold, the global error budget, and the anchor point set to form an error control parameter table.

[0045] Specifically, the global error budget ε total Based on the application scenario, this embodiment uses 0.1% to 0.5% of the model bounding box diagonal length; the error budget allocation adopts a weighted allocation strategy, and for windows marked as periodic segments, the time error budget ε t The proportion is 30%, and the spatial error budget ε s The percentage is 70%, while for non-periodic segments it is 50% and 50% respectively; the integer step size Δ of the coordinates is calculated using the formula Δ=ε. s / (2 b -1) Calculate, where ε s For the spatial error budget, b is the number of bits represented by an integer, which is set to 16 bits by default; the time quantization threshold τ is also determined by the formula τ=ε. t / (2 b-1) Calculation; Anchor point detection is achieved by calculating the rate of change acceleration (rate of change of rate of change) between adjacent time points. When the acceleration exceeds three times the standard deviation of the rate of change, the time point is marked as an anchor point; The final error control parameter table is organized in the form of a structure, containing the field {ε total ,ε s ,ε t ,Δx,Δy,Δz,τ,{a ps}}, where Δx, Δy, and Δz correspond to the integer step sizes in the three coordinate axis directions, {a ps} represents the set of anchor points.

[0046] Specifically, the top N frequency components with the highest energy percentage are identified as the target frequency components.

[0047] It should be noted that the value of N can be dynamically determined according to application requirements. For scenarios with high precision requirements, it can be set to 5-8, and for scenarios with lower precision requirements, it can be set to 2-3. The recognition process first arranges the spectral energy in descending order, and then accumulates the energy from high to low until it reaches 95% of the total energy. The frequency components included in these accumulation processes are determined as target frequency components. For special cases such as white noise signals, if the accumulated energy of the first 10 frequency components still does not reach 70% of the total energy, it is judged to have no obvious periodic characteristics and is forcibly marked as a non-periodic segment.

[0048] Specifically, time-series data is obtained by arranging the three-dimensional coordinates and attribute values ​​of each vertex in chronological order within a fixed window, forming multiple parallel one-dimensional time series. For terrain change parameter data, the elevation change values, displacement vectors, and scale parameters of each terrain feature point are converted into raster matrices or grids and then arranged in chronological order to form a time series of terrain parameters. For complex attributes such as texture coordinates, two-dimensional UV values ​​are first mapped to one-dimensional scalars before constructing the time series. These time-series data serve as inputs to Fourier transforms and prediction models, forming the basic data structure for time coding. Each time series is associated with a specific spatial point or terrain node to capture the changing features of the temporal dimension while preserving spatial semantics.

[0049] It should be noted that, in this invention, terrain nodes refer to key points representing terrain features in a digital elevation model (DEM) or digital terrain model (DTM), including but not limited to hilltops, rift points, runoff nodes, gully head points, and saddle points. As control points for changes in the terrain surface, terrain nodes possess specific spatial locations and elevation attributes, effectively describing the geometric features and structural relationships of the terrain. In the aforementioned spatiotemporal data compression processing, the elevation change value of a terrain node represents the vertical displacement change of the node, the displacement vector represents the movement of the node on the horizontal plane, and the scale parameter represents the scaling change of the terrain features.

[0050] In one embodiment, generating a time-coded stream based on a spatiotemporal block and an error control parameter table includes: determining the type of the current spatiotemporal block based on the window label; for spatiotemporal blocks marked as periodic segments, extracting the amplitude and phase parameters of the target frequency components through frequency domain transformation to form a frequency domain parameter set; for spatiotemporal blocks marked as non-periodic segments, constructing a time series prediction model to calculate predicted values ​​and comparing them with actual values ​​to obtain a residual sequence; performing dead-zone quantization on the frequency domain parameter set and the residual sequence according to a time quantization threshold to control quantization error; retaining the original data at the corresponding time points based on the anchor point set without quantization processing; and organizing the quantized frequency domain parameter set and the residual sequence into a time-coded stream.

[0051] Specifically, for spatiotemporal blocks marked as periodic segments, the frequency domain transformation is implemented using the Fast Fourier Transform (FFT) algorithm. The x, y, and z coordinates of each vertex within the spatiotemporal block are treated as independent signals for transformation. Based on the previously identified N frequency components with the highest energy proportions, their amplitude A and phase φ parameters are extracted. Each frequency component fi is represented as a binary tuple <Ai,φi>. All frequency components form the frequency domain parameter set {<f1,A1,φ1>,<f2,A2,φ2>,...,<fN,AN,φN>}. The time series prediction model uses an autoregressive moving average (ARMA) model. For non-periodic spatiotemporal blocks, The parameters of the ARMA model are estimated using the least squares method. The model order p and q are automatically selected to their optimal values ​​according to the Akaike Information Criterion (AIC). In this embodiment, p∈[1,5] and q∈[1,3] are chosen. The difference between the predicted and actual values ​​forms a residual sequence. A centralized dead-zone quantizer is used for dead-zone quantization with a dead-zone width of 2τ and a quantization step size of τ. When the absolute value of the residual or frequency domain parameter is less than τ, it is quantized to zero; when it is greater than τ, it is uniformly quantized according to the quantization step size τ. For time points in the anchor point set, the full-precision representation of the original data is directly saved to ensure accurate reconstruction of keyframes. The final time-coded stream uses <type,params,q>. data ,{aps}> represents, where type indicates the coding type (0 for frequency domain parameters, 1 for prediction residuals), params are the coding parameters, and q data {aps} is the set of frequency domain parameters or residual sequence after quantization, and {aps} is the set of anchor points.

[0052] In one embodiment, integerizing standardized data entries and generating a spatial encoded stream using double difference and run-length encoding includes: quantizing the three-dimensional coordinates and attribute values ​​of standardized data entries according to the coordinate integerization step size to obtain an integer sequence; performing first-order difference calculation on the integer sequence to obtain a first-order difference sequence, and performing second-order difference calculation to obtain a second-order difference sequence; analyzing the sparsity characteristics of the second-order difference sequence to determine continuous repeating values; encoding the continuous repeating values ​​using run-length encoding to obtain a run-length encoded sequence; and further compressing the run-length encoded sequence by entropy encoding to generate a spatial encoded stream.

[0053] Specifically, in this embodiment, the coordinate integerization step size is obtained through the aforementioned Δ formula; the integerization process for the three-dimensional coordinates (x, y, z) is x int =round(x / Δx), y int =round(y / Δy), z int =round(z / Δz), where Δx, Δy, and Δz are the integer step sizes for each coordinate axis; the integerization of attribute value a uses a linear mapping, first normalizing the attribute value to the [0,1] interval, and then mapping it to [0,2]. ba The integer range is [-1], where ba is the number of bits for attribute encoding, set to 8 bits in this embodiment; the first-order difference sequence d1 is obtained by calculating the difference between adjacent integer values, d1[i]=v[i]-v[i-1], where v is an integer sequence; the second-order difference sequence d2 is obtained by calculating the difference between the first-order difference sequences, d2[i]=d1[i]-d1[i-1]; the run-length encoding adopts the form of <value,length> tuples, indicating that the value value appears consecutively length times, and non-zero values ​​in the second-order difference sequence are directly retained; in this invention, the entropy encoding adopts an adaptive arithmetic encoding algorithm, which automatically optimizes the encoding efficiency according to the data distribution characteristics by dynamically adjusting the probability model, and the final generated spatial encoding stream is in <int p diff d ,rle d > indicates that int p For integer-based parameters, diff d For differential coding parameters, rle d This is the run-length encoded data.

[0054] In one embodiment, edge-folding lightweighting and texture compression are performed on grid data in standardized data entries to generate a joint encoding result. This includes: calculating a quadratic error metric for each edge of the grid data; selecting edges for folding operations according to the magnitude of the error metric to obtain a simplified grid; using a spatial error budget as a distance constraint parameter, calculating the maximum distance between the current simplified grid and the original grid during the folding process; stopping folding when the maximum distance reaches the constraint parameter to obtain simplified grid structure data; constructing a two-dimensional mapping for the texture information of the grid data; calculating the pixel occupancy rate of each region; merging texture regions with occupancy rates below a preset threshold into continuous regions to form an optimized texture atlas; applying a block compression algorithm to the optimized texture atlas to generate compressed texture data; organizing the simplified grid structure data and compressed texture data into a geometric texture lightweight stream; and combining the temporal encoding stream, spatial encoding stream, and geometric texture lightweight stream into a joint encoding result.

[0055] Specifically, the quadratic error metric in edge folding lightweighting is obtained by calculating the quadratic error matrices Q1 and Q2 of the two vertices v1 and v2 of the edge, and the combined error is v. T (Q1+Q2)v, where v is the 4×1 homogeneous coordinate vector of the new vertex after folding, represented as v=[x,y,z,1] T The new vertex position is obtained by solving (Q1+Q2)v'=[0,0,0,1]. T We obtain v', where v' is the optimized homogeneous coordinate vector. If the upper 3×3 submatrix of matrix Q1+Q2 is not invertible, the midpoint of the edge is taken as the new vertex position. The folding operation sequence arranges all edges in ascending order according to the quadratic error metric. Each time, the edge at the top of the heap is selected for folding, and the error metric of the affected edge and the heap position are updated. The distance constraint parameter adopts the Hausdorff distance metric. The distance calculation is accelerated by constructing an octree spatial index structure. After every 10 folding operations, the Hausdorff distance between the current simplified mesh and the original mesh is checked. Folding stops when the distance exceeds the spatial error budget εs. Texture region occupancy The utilization rate is calculated using the pixel density map in the UV space. The UV space is divided into a 64×64 grid, and the proportion of texture pixels in each grid is counted. Areas with a utilization rate of less than 25% are considered low-utilization areas. Texture atlas optimization uses a rectangular binning algorithm, keeping high-utilization areas unchanged and rearranging and merging low-utilization areas to minimize the total area of ​​the texture atlas. The block compression algorithm adaptively selects either DXT1 (without alpha channel) or DXT5 (with alpha channel) format based on texture characteristics, and performs 4×4 block encoding on the RGB color channels and alpha channels respectively. The lightweight geometric texture stream uses <mesh>. d ,tex d ,uv m > indicates that the meshd To simplify mesh data, including vertex coordinates, connectivity, and attributes, tex d To compress texture data, UV m This represents the UV mapping relationship.

[0056] It should be noted that edge-folding lightweight algorithm is one of the core technologies for real-world 3D model compression. Real-world 3D models consist of elements such as vertices, faces, and textures. These vertices are key nodes in the model, reflecting the details of the terrain outline. By connecting three adjacent vertices, triangular faces are formed, creating an irregular triangular mesh model. The edge-folding algorithm used in this invention is based on Quadratic Error Metric (QEM), which can intelligently preserve the key geometric features of the model while effectively reducing the amount of data.

[0057] In the construction of realistic 3D environments, lightweighting of 3D models is a pressing need. Among model simplification algorithms, the Quadratic Error Metric (QEM) algorithm is considered a classic method, but traditional QEM algorithms lack a dedicated mechanism to preserve key details during the simplification process. This invention effectively solves this problem by introducing Hausdorff distance as a spatial error constraint parameter. Hausdorff distance can accurately measure the maximum deviation between the simplified mesh and the original mesh, ensuring that key geometric features are preserved during model simplification.

[0058] For the unique "model nodes + triangles + textures" organization of realistic 3D models, this invention not only focuses on simplifying the mesh structure but also optimizes and compresses the texture data. Through pixel density analysis in UV space and a rectangular binning algorithm, underutilized texture regions are rearranged, and a block compression algorithm is applied to achieve efficient texture data compression while maintaining the model's visual quality. This compression method, which combines geometry and texture, is particularly suitable for the data characteristics of realistic 3D models, significantly reducing data volume and improving transmission and rendering efficiency while ensuring visual quality.

[0059] In one embodiment, based on the joint coding results, compressed data is divided into a base layer and several enhancement layers, and an associative hierarchical storage structure and a block-level index table are established to generate streamable compressed data. This includes: extracting data from the temporal-coded stream, spatial-coded stream, and lightweight geometric texture stream that satisfy the spatial error budget and temporal error budget, forming the base layer; organizing the unextracted data from the temporal-coded stream, spatial-coded stream, and lightweight geometric texture stream into several enhancement layers; establishing a temporal-spatial associative hierarchical storage structure based on the window, spatial neighborhood range, and window label of the spatiotemporal block to record the correspondence between each layer and the spatiotemporal block; establishing a block-level index table based on the associative hierarchical storage structure to record the index information of the window, spatial neighborhood range, and window label corresponding to each layer; storing simplified mesh structure data only once in the base layer for recurring mesh data, and storing reference identifiers pointing to the base layer in the enhancement layers; and encapsulating the base layer and enhancement layers using the block-level index table to generate streamable compressed data.

[0060] Specifically, in this invention, the division between the base layer and the enhancement layer adopts a hierarchical error control strategy. The base layer contains data that satisfies 1 times the original error budget, the first enhancement layer contains additional data that satisfies 0.5 times the original error budget, the second enhancement layer contains additional data that satisfies 0.25 times the original error budget, and so on. For the temporally encoded stream, the base layer contains the top N / 2 frequency component parameters with the highest energy proportion, or the main coefficients and larger residual values ​​of the ARMA model. For the spatially encoded stream, the base layer contains a low-precision representation of the coordinates (in this embodiment, 2 / 3 of the total number of bits is used). For the lightweight geometric texture stream, the base layer contains a simplified mesh with a 50% reduction in face size and a texture with a reduced resolution by one level. The associative hierarchical storage structure is implemented using a tree structure, where the root node represents the base layer, the child nodes represent the enhancement layers, and the edges between nodes represent hierarchical dependencies. The block-level index table uses a hash table structure, with the key being <w id ,s id The `<ref>` tag combination specifies the offset and length of each layer's data. For duplicate references to grid data, a reference counting mechanism is used. When a grid topology similarity exceeding 95% is detected, only the transformation matrix and reference identifier are stored. The reference identifier includes `<ref>` tags. id ,transform> tuple,ref id The data points to the base layer's grid data, where transform is the affine transformation matrix from the reference grid to the current grid. The final compressed data packet uses a chunked encapsulation format. The header contains global metadata, a block-level index table, and a description of hierarchical relationships. The data portion is organized into blocks and supports random access. Each data block contains a block header and block data. The block header indicates the block type, level, and dependencies, allowing the decoding end to selectively load and decode data at the appropriate level based on device performance and bandwidth conditions.

[0061] To facilitate understanding of the above-mentioned technical solution of the present invention, the following detailed description uses a terrain data processing system of a geographic information system development company as an example:

[0062] When developing a multi-platform real-scene 3D geographic information visualization system, the company faced the problem of slow loading and rendering lag of large-scale real-scene 3D model data on mobile devices. In particular, the high-precision real-scene 3D model of urban buildings (approximately 500,000 vertices and 1 million triangles) and its multi-period time-series data, in their uncompressed state, a single urban real-scene model (containing 240 building models) occupied 3.2GB of storage space, which seriously affected the running performance of the application on low- and mid-range mobile devices.

[0063] To address this issue, the development team implemented the spatiotemporal data compression method of this invention. First, the 3D model data was preprocessed to establish standardized data entries and unify the coordinate system and time axis (collected quarterly). Then, dynamic windowing was performed based on the model's geometric change rate. The urban building change sequence was divided into three windows: two quarterly windows for rapidly developing areas (rapid change segments) and six quarterly windows for stable development areas (slow change segments). The frequency domain characteristics of each window were analyzed using FFT transformation, marking the first and third windows as aperiodic segments and the second window as aperiodic segment (exhibiting obvious seasonal variation characteristics). An error control parameter table was established based on the global error budget (set as 0.2% of the model bounding box diagonal), and the coordinate data was processed using 16-bit integers. The amplitude and phase parameters of the four main frequency components were extracted for the periodic segments, and the residuals were calculated using an ARMA model for the aperiodic segments. The real-world 3D model simplified the number of faces by 35% through an edge-folding algorithm, and the Hausdorff distance constraint ensured the preservation of key geometric features. The model texture size was reduced by 40% through UV space optimization and block compression algorithms.

[0064] The final compressed data consists of a base layer (450MB) and two enhancement layers (totaling 180MB), achieving an overall compression rate of 80%. It can achieve real-time rendering at 25 frames per second on mid-range mobile devices while maintaining high-fidelity visual quality; the difference from the original realistic 3D model is virtually imperceptible to the naked eye. This system has been successfully applied to multiple geographic information system products within the company, effectively improving the image processor's ability to efficiently compress and process large-scale realistic 3D spatiotemporal data in real time, and solving the problem of low image data compression performance for realistic 3D terrain models on multiple terminal devices.

[0065] like Figure 3 As shown, according to another embodiment of the present invention, a spatiotemporal data compression system based on lightweight processing is also provided, the spatiotemporal data compression system based on lightweight processing includes:

[0066] Preprocessing module 1 is used to establish standardized data entries based on the spatiotemporal data to be compressed, form a window set by segmenting the time axis, periodically distinguish the windows according to the frequency domain energy distribution, construct spatiotemporal blocks, and establish an error control parameter table based on the global error budget.

[0067] Encoding generation module 2 is used to generate a temporal encoding stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial encoding stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results.

[0068] The layered encapsulation module 3 is used to divide the compressed data into a basic layer and several enhancement layers based on the joint encoding results, and to establish an associative layered storage structure and a block-level index table to generate compressed data that can be streamed.

[0069] 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 spatiotemporal data compression method based on lightweight processing, characterized in that, include: S1. Based on the spatiotemporal data to be compressed, standardized data entries are established through preprocessing. The rate of change of the three-dimensional coordinates in the standardized data entries between adjacent time points is calculated. Based on the adaptive window mechanism of the rate of change, a window set is formed by segmenting the time axis. The windows are periodically judged according to the frequency domain energy distribution to construct spatiotemporal blocks. An error control parameter table is established based on the global error budget. S2. Generate a time-coded stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial-coded stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results. S3. Based on the joint coding results, the compressed data is divided into a basic layer and several enhancement layers. An associative hierarchical storage structure and a block-level index table are established to generate compressed data that can be streamed. The generation of the time-coded stream based on the spatiotemporal block and error control parameter table includes: Determine the type of the current spacetime block based on the window label; For a spatiotemporal block marked as a periodic segment, the amplitude and phase parameters of the target frequency component are extracted through frequency domain transformation to form a frequency domain parameter set; For a spatiotemporal block marked as a non-periodic segment, a time series prediction model is constructed to calculate the predicted value, and the residual sequence is obtained by comparing it with the actual value. The frequency domain parameter set and residual sequence are quantized with dead zone according to the time quantization threshold to control quantization error; based on the anchor point set, the original data at the corresponding time point is retained without quantization processing. The quantized frequency domain parameter set and residual sequence are organized into a time-coded stream.

2. The spatiotemporal data compression method based on lightweight processing according to claim 1, characterized in that, The spatiotemporal data to be compressed includes at least one of vertex data, coordinate data, normal data, color data, texture data, terrain change parameter data, and trajectory point data; Establishing standardized data entries through preprocessing includes: The spatiotemporal data to be compressed is identified by data type to determine the data elements it contains; The data elements are uniformly transformed to the same coordinate system and time axis, and outliers exceeding the preset threshold range are removed to obtain the initial data; The initial data sequence is processed to unify the sampling rate, and various attributes are aligned to obtain aligned data; The aligned data is organized into standardized data entries, which include time points, three-dimensional coordinates, attribute values, and grid data.

3. The spatiotemporal data compression method based on lightweight processing according to claim 2, characterized in that, An adaptive windowing mechanism based on the rate of change, which uses time axis segmentation to form a window set, includes: The dynamic window length is set according to the rate of change. For regions with a rate of change higher than the first preset threshold, a window with a length smaller than the preset length is used, and for regions with a rate of change lower than the second preset threshold, a window with a length greater than the preset length is used. A window set is generated by segmenting the timeline using a sliding window with a dynamic window length. Create a window-segment index for each window in the window set to facilitate block-based processing of standardized data entries.

4. The spatiotemporal data compression method based on lightweight processing according to claim 1, characterized in that, Based on the frequency domain energy distribution, the window is periodically discriminated, and the spatiotemporal block is constructed as follows: Perform Fourier transform on the time series data of the standardized data entries within each window in the window set to calculate the frequency domain energy distribution; The energy proportion of the target frequency component in the frequency domain energy distribution is determined based on the spectral threshold. When the proportion exceeds the spectral threshold, the window is marked as a periodic segment; otherwise, it is marked as a non-periodic segment, thus obtaining the window label. Within each window of the window set, a spatial neighborhood is defined, which is the clustered area of ​​three-dimensional coordinates in adjacent standardized data entries; Window, spatial neighborhood range, and window label are combined to form a spatiotemporal block, which serves as the basic unit for compression processing.

5. The spatiotemporal data compression method based on lightweight processing according to claim 4, characterized in that, The error control parameter table established based on the global error budget includes: Define a global error budget to represent the maximum allowable error after compressed reconstruction; Based on the window label and spatial neighborhood range, the global error budget is allocated into a spatial error budget and a temporal error budget; Determine the coordinate integerization step size based on spatial error budgeting; Determine the time quantization threshold based on the time error budget; The rate of change of three-dimensional coordinates and attribute values ​​in standardized data entries is detected, and time points where the rate of change exceeds a preset mutation threshold are marked as anchor points to generate an anchor point set. The coordinate integerization step size, time quantization threshold, global error budget, and anchor point set are combined to form an error control parameter table.

6. The spatiotemporal data compression method based on lightweight processing according to claim 1, characterized in that, Integerizing standardized data entries and generating a spatial encoded stream using double difference and run-length encoding includes: The three-dimensional coordinates and attribute values ​​of standardized data entries are quantized according to the coordinate integerization step size to obtain an integer sequence; A first-order difference sequence is obtained by performing a first-order difference calculation on an integer sequence, and a second-order difference sequence is obtained by performing a second-order difference calculation on the first-order difference sequence. Analyze the sparsity characteristics of the second-order difference sequence to determine the continuous repeating values; Run-length encoding is used to encode consecutive repeating values ​​to obtain a run-length encoded sequence; A spatial coded stream is generated by further compressing the run-length encoded sequence by entropy encoding.

7. The spatiotemporal data compression method based on lightweight processing according to claim 1, characterized in that, Edge-folding and texture compression are performed on the grid data in the standardized data entries to generate joint encoding results, including: For the grid data, calculate the quadratic error metric for each edge, and select the edges for folding operation in order of the magnitude of the error metric to obtain a simplified grid; Using spatial error budget as a distance constraint parameter, the maximum distance between the current simplified mesh and the original mesh is calculated during the folding process. Folding stops when the maximum distance reaches the constraint parameter, and the simplified mesh structure data is obtained. A two-dimensional mapping is constructed on the texture information of the mesh data, the pixel occupancy rate of each region is calculated, and texture regions with occupancy rates below a preset threshold are merged into continuous regions to form an optimized texture atlas. The optimized texture atlas is encoded using a block compression algorithm to generate compressed texture data. Simplified mesh structure data and compressed texture data are organized into a lightweight geometric texture stream; The temporal coding stream, spatial coding stream, and lightweight geometric texture stream are combined into a joint coding result.

8. The spatiotemporal data compression method based on lightweight processing according to claim 1, characterized in that, Based on the joint coding results, the compressed data is divided into a basic layer and several enhancement layers. An associative hierarchical storage structure and a block-level index table are established to generate streamable compressed data, including: Based on the joint coding results, the data that satisfy the spatial error budget and the temporal error budget, the temporal coding stream, the spatial coding stream, and the lightweight geometric texture stream are extracted to form the base layer; Unextracted data from the temporally encoded stream, spatially encoded stream, and lightweight geometric texture stream are organized into several enhancement layers; Based on the window, spatial neighborhood range, and window label of the spatiotemporal block, a time-space associative hierarchical storage structure is established to record the correspondence between each layer and the spatiotemporal block. A block-level index table is established based on the associative hierarchical storage structure to record the index information of the window, spatial neighborhood range and window label corresponding to each layer; For recurring grid data, the simplified grid structure data is stored only once in the base layer, and a reference identifier pointing to the base layer is stored in the enhancement layer. The base layer and enhancement layer are encapsulated using a block-level index table to generate compressed data that can be streamed.

9. A spatiotemporal data compression system based on lightweight processing, used to implement the spatiotemporal data compression method based on lightweight processing according to any one of claims 1-8, characterized in that, The system includes: The preprocessing module is used to establish standardized data entries based on the spatiotemporal data to be compressed, calculate the rate of change of the three-dimensional coordinates in the standardized data entries between adjacent time points, form a window set by segmenting the time axis based on the rate of change, periodically distinguish the windows according to the frequency domain energy distribution, construct spatiotemporal blocks, and establish an error control parameter table based on the global error budget. The encoding generation module is used to generate a temporal encoding stream based on the spatiotemporal block and error control parameter table, integerize the three-dimensional coordinates and attribute values ​​in the standardized data entries, generate a spatial encoding stream through double difference and run-length encoding, perform edge folding and lightweighting and texture compression on the mesh data in the standardized data entries, and generate joint encoding results. The layered encapsulation module is used to divide compressed data into a base layer and several enhancement layers based on the joint encoding results, and to establish an associative layered storage structure and block-level index table to generate compressed data that can be streamed. The generation of the time-coded stream based on the spatiotemporal block and error control parameter table includes: Determine the type of the current spacetime block based on the window label; For a spatiotemporal block marked as a periodic segment, the amplitude and phase parameters of the target frequency component are extracted through frequency domain transformation to form a frequency domain parameter set; For a spatiotemporal block marked as a non-periodic segment, a time series prediction model is constructed to calculate the predicted value, and the residual sequence is obtained by comparing it with the actual value. The frequency domain parameter set and residual sequence are quantized with dead zone according to the time quantization threshold to control quantization error; based on the anchor point set, the original data at the corresponding time point is retained without quantization processing. The quantized frequency domain parameter set and residual sequence are organized into a time-coded stream.

Images