Trajectory compression method, system and storage medium for multi-modal attributes of sea-air moving targets

Abstract

The application discloses a kind of trajectory compression methods, systems and storage medium for sea-air moving target multi-modal attribute, the method includes receiving original trajectory data stream, parsing and extracting multi-modal attribute, attribute is classified as core space-time attribute, low-frequency change attribute and high-frequency change attribute;Extract key node based on preset condition, record its node index, core space-time attribute, low-frequency and high-frequency change attribute, encode its low-frequency change attribute;Only record the low-frequency attribute of non-key node and encode, calculate and predict the residual of its position information value and quantize;Encapsulate and entropy coding output according to hierarchical structure to the data processed.This application can maintain sea-air moving target trajectory space-time semantic information integrity under high compression ratio and support compressed domain query.

Description

Technical Field

[0001] This invention relates to data compression, and more particularly to a trajectory compression method, system, and storage medium for multimodal attributes of moving targets at sea and in the air. Background Technology

[0002] With the development of information systems, massive amounts of trajectory data for air and sea targets such as civil flights and commercial vessels are generated every day. This data not only contains high-precision spatiotemporal location information, such as longitude, latitude, altitude, and timestamps, but also includes rich multimodal attribute information, such as speed, heading, platform name, platform type, and current mission status.

[0003] Most existing trajectory data compression techniques focus on compressing pure spatial position sequences, such as using the Douglas-Peucker algorithm to reduce the number of trajectory points, or performing linear fitting based on physical quantities such as velocity and acceleration. However, these methods have the following drawbacks: (1) They usually discard or process attribute data independently of spatiotemporal information, resulting in the compressed data losing key contextual information and failing to restore the complete behavioral profile of the target; (2) After independent compression, the strict correspondence between attributes and spatiotemporal points may be destroyed, for example, it is impossible to accurately determine whether the platform's task status has changed at a certain key turning point; (3) Simply using a general compression algorithm on attribute data cannot utilize its semantic relevance and temporal continuity, resulting in a low compression ratio; (4) When compressed data is used for subsequent spatiotemporal queries and analysis, it often needs to be completely decompressed, resulting in low processing efficiency. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a trajectory compression method, system, and storage medium for multimodal attributes of sea and air moving targets that can maintain the integrity of spatiotemporal semantic information of the trajectory of sea and air moving targets at a high compression ratio and support compression domain query.

[0005] Technical solution: The trajectory compression method for multimodal attributes of moving sea and air targets described in this invention includes:

[0006] Receive the raw trajectory data stream, parse and extract the multimodal attributes, and classify the multimodal attributes into core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes;

[0007] Based on preset conditions, key nodes are extracted from the original trajectory data stream, and the node index, core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes of the key nodes are recorded. The low-frequency change attributes are then encoded.

[0008] For non-critical nodes between any adjacent critical nodes, only the low-frequency change attributes of the non-critical nodes are recorded and encoded; for the location information in the core spatiotemporal attributes of the non-critical nodes, the residual between the predicted value and the location information obtained based on the previous critical node is calculated and the residual is quantified.

[0009] The data recorded and processed by critical and non-critical nodes are encapsulated in a hierarchical structure, and the final compressed output is generated by applying an entropy coding algorithm.

[0010] By distinguishing the multimodal attributes of trajectory data into core spatiotemporal attributes, low-frequency change attributes, and high-frequency change attributes, low-frequency change attributes are generally stable in the time domain. Only the full value is recorded at key nodes, while only the changed portion is recorded at non-key nodes, effectively eliminating the overhead of redundant storage of static attributes across a large number of continuous trajectory points. High-frequency change attributes retain only the original values ​​of key nodes, and data from non-key nodes can be recovered through the model, avoiding redundancy caused by point-by-point storage. Simultaneously, the location information in the core spatiotemporal attributes of non-key nodes is calculated and quantized using residuals based on linear prediction. Only a much smaller amount of data needs to be stored than the quantized residuals of the complete location information, allowing for the subsequent recovery of the complete location information of non-key nodes. Since the number of non-key nodes is significantly larger than that of key nodes, this method can substantially reduce the amount of data stored. The synergistic effect of these steps enables the compressed output to represent the original trajectory with an extremely high compression ratio. Furthermore, because key nodes retain complete multimodal attribute records, and non-key node change attributes are encoded sequentially, the compressed data fully preserves spatiotemporal semantic information, and can be directly used for behavioral analysis or semantic querying after decompression. Furthermore, the hierarchical encapsulation combined with entropy coding further eliminates statistical redundancy within the data packets and gives the compressed stream itself structured characteristics, which is beneficial for fast retrieval without decompressing all the data.

[0011] Preferably, the core spatiotemporal attributes include timestamp, longitude, latitude, and altitude; low-frequency change attributes include country, platform name, platform type, and mission; and high-frequency change attributes include speed and heading.

[0012] By clearly defining the specific categories of attributes, a clear basis is provided for implementing differentiated coding for attributes with different change frequencies, ensuring the pertinence and efficiency of the compression strategy.

[0013] As a preferred option, the preset conditions for extracting key nodes are: the spatial Euclidean distance, rate of change of velocity, and heading deflection angle of the previous key node all exceed the corresponding preset thresholds; the preset thresholds used to filter key nodes are adjusted according to the task type.

[0014] By comprehensively utilizing spatial distance, rate of change of velocity, heading deflection angle, and task type as key node selection criteria, the extracted key nodes can more accurately capture the target's motion characteristics and task semantic changes, effectively maintaining the trajectory's shape and semantic fidelity while reducing the amount of data.

[0015] As a preferred approach, a global dictionary table is created for low-frequency change attributes, and all low-frequency change data of critical nodes are initially encoded based on this dictionary table. For non-critical nodes, the low-frequency change attribute values ​​are only initially encoded using the global dictionary table when they change compared to the preceding critical nodes. After the initial encoding is completed, the sliding window dictionary encoding method is used to encode all the initial encoded values ​​a second time.

[0016] By creating a global dictionary table of low-frequency attributes to perform preliminary encoding on critical and non-critical nodes, the amount of data stored can be significantly reduced. After the preliminary encoding, a sliding window dictionary encoding is used to perform secondary encoding on the preliminary encoded values, which can further compress the amount of data and further reduce the final data storage footprint.

[0017] As a preferred method, for non-critical nodes, the predicted value of the location information is obtained by linear interpolation based on the timestamps of the previous critical node and the non-critical node; the residual is calculated by the difference between the actual location information of the non-critical node and the predicted value, and the residual is quantified.

[0018] By generating predicted values ​​and calculating residuals through timestamp-based linear interpolation, linear redundancy in core spatiotemporal data between adjacent key nodes is effectively eliminated, which significantly reduces the numerical range and storage overhead of subsequent quantization.

[0019] As a preferred method, the operation of quantizing the residual is to quantize the calculated floating-point residual value into an integer with a preset number of bits.

[0020] By lossily quantizing floating-point residuals into integers, the storage volume of core spatiotemporal data is further significantly reduced while keeping information loss under control.

[0021] Preferably, the hierarchical structure includes a file header and a data body. The file header includes a dictionary table created for low-frequency change attributes and complete information about the initial key nodes.

[0022] By placing the dictionary table and initial key node information at the beginning of the file, global reference information can be quickly obtained during decompression, supporting fast initialization and conditional queries without full decompression.

[0023] Preferably, the data body alternately stores key node data blocks and differential data blocks in chronological order. The differential data blocks contain quantized residual vectors of non-key nodes between adjacent key nodes and encoded data of low-frequency change attributes.

[0024] By alternating between critical node data blocks and differential data blocks in chronological order, it is ensured that the data of critical nodes and non-critical nodes between adjacent critical nodes can be stored in an orderly manner according to time. When querying data later, it is not necessary to parse the entire data block; only the corresponding data block needs to be parsed according to time for fast querying.

[0025] The trajectory compression system for multimodal attributes of moving sea and air targets as described in this invention includes:

[0026] Data preprocessing module: used to receive raw trajectory data stream, parse and extract multimodal attributes, and classify multimodal attributes into core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes;

[0027] Key node analysis module: used to extract key nodes from the original trajectory data stream based on preset conditions, and record the node index, core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes of the key nodes;

[0028] Attribute compression module: used to encode low-frequency change attributes of key nodes; for non-key nodes between any adjacent key nodes, only the low-frequency change attributes of non-key nodes that have changed are encoded; for the location information in the core spatiotemporal attributes of non-key nodes, the residual between the location information and the predicted value based on the location information obtained from the previous key node is calculated, and the residual is quantized.

[0029] Data encapsulation module: This module encapsulates the data recorded and processed by key nodes and non-key nodes in a hierarchical structure, and applies an entropy coding algorithm to generate the final compressed output.

[0030] The present invention provides a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions that, when executed by a computing device, cause the computing device to perform the method described in any of the preceding claims.

[0031] Beneficial effects: By organically combining techniques such as key node extraction, hierarchical encoding of multimodal attributes, and residual quantization of non-key nodes, integrated high-fidelity compression of sea and air motion target trajectory data is achieved. While obtaining a high compression ratio, the semantic correlation between spatiotemporal information and multimodal attributes is fully preserved. The decompressed data can be directly used for semantic analysis and fast querying, and the algorithm has low complexity, making it suitable for real-time processing scenarios. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall process of the method described in this invention;

[0033] Figure 2 This is a schematic diagram of the key node extraction results;

[0034] Figure 3 This is a schematic diagram of a hierarchical data packet structure. Detailed Implementation

[0035] like Figure 1 As shown, the trajectory compression method for multimodal attributes of moving sea and air targets according to the present invention includes the following process:

[0036] Step S1: Data preprocessing.

[0037] The system receives a raw trajectory data stream, consisting of a series of trajectory points sorted by timestamps. It parses and extracts the multimodal attributes from each trajectory point, then sorts these attributes to form an attribute array, which includes attribute indices and values. Specifically, the attributes are categorized into three types:

[0038] (1) Core spatiotemporal attributes, including: timestamp, longitude, latitude and altitude.

[0039] (2) Low-frequency change attributes, including: country, platform name, platform type and task.

[0040] (3) High-frequency change attributes, including speed and heading.

[0041] Step S2: Extraction of key nodes.

[0042] Based on preset conditions, the core spatiotemporal sequence is processed to extract key nodes from the original trajectory data stream. The preset conditions for extracting key nodes are: the spatial Euclidean distance (considering altitude) to the previous key node exceeds a preset threshold; the rate of change of velocity and the heading deflection angle both exceed their corresponding preset thresholds; and the preset thresholds used to filter key nodes are adjusted according to the mission type. These preset thresholds are determined empirically based on air and sea targets, such as a distance exceeding 100 kilometers from the previous key point, a velocity change exceeding 10 kilometers per hour, or a heading deflection angle exceeding 15 degrees. For example, for search and rescue missions, by adjusting the corresponding preset thresholds, hovering nodes and nodes with altitude changes are forcibly retained, while nodes in straight cruising segments can be discarded; for cruising missions, by adjusting the corresponding preset thresholds, turning nodes, low-speed nodes, and area boundary nodes are forcibly retained, while nodes in repetitive segments of fixed routes can be discarded.

[0043] like Figure 2 As shown (green dots in the figure represent the selected key nodes), for the trajectory of a platform performing a search and rescue mission, the above method was used to identify turning points and other points of abrupt changes in motion as key nodes. For each selected key node, the system records its node index and the complete original values ​​of its core spatiotemporal attributes.

[0044] Step S3: Multimodal attribute hierarchical encoding.

[0045] This step differentiates between low-frequency and high-frequency change attributes.

[0046] S3.1: Low-frequency change attribute coding.

[0047] A global dictionary table is created for the low-frequency changing attributes of trajectory objects. For enumerated data, ordinary custom numbers are used as the encoding for the corresponding attributes in the global dictionary table; for character array data, their hash values ​​are used. The global dictionary table is used for initial encoding of nodes: for critical nodes, the encoded values ​​corresponding to all low-frequency changing attribute values ​​are recorded in a fixed order according to their attribute indices. For non-critical nodes, the attribute index and the corresponding encoding of the changed attribute are recorded only when the low-frequency changing attribute value changes compared to a preceding critical node. After the initial encoding is completed, a sliding window dictionary encoding method is used to perform secondary encoding on all the above initial encodings to further compress the data volume.

[0048] S3.2: High-frequency attribute changes.

[0049] For frequently changing attributes such as speed and heading, linear interpolation or second-order dynamic models are used to fit the changes in speed and heading between adjacent critical nodes. During data storage, the original speed and heading attribute values ​​of critical nodes are recorded. The speed and heading attributes of non-critical nodes are not stored and can be recovered from the model during decompression.

[0050] Step S4: Non-critical node residual coding and quantization.

[0051] For the location information (longitude, latitude, and altitude) in the core spatiotemporal attributes of all non-critical nodes, the residuals between these residuals and the predicted values ​​obtained through linear interpolation based on the previous critical node are calculated. Specifically, linear interpolation is performed using the timestamps of the previous critical node and the non-critical node to obtain the predicted location value of the non-critical node; the difference between the actual location information and the predicted value is calculated to form a residual vector (Δlongitude, Δlatitude, Δaltitude). Next, this residual vector is subjected to lossy quantization, reducing the floating-point precision to a preset number of bits, for example, quantizing to a 16-bit integer.

[0052] Step S5: Data encapsulation and entropy coding.

[0053] like Figure 3As shown, the data processed in the preceding steps is organized into a layered data packet structure. The layered structure consists of a file header and a data body. The file header includes a dictionary table created for low-frequency changing attributes and complete information about the initial critical nodes. The data body includes critical node data blocks and differential data blocks stored alternately in time series. The critical node data blocks contain complete information about that point, while the differential data blocks contain the quantized residual vectors of non-critical nodes between adjacent critical nodes and the encoded data of low-frequency changing attributes. Finally, a lossless entropy coding algorithm, such as arithmetic coding or interval coding, is applied to the entire layered data packet to further eliminate statistical redundancy and generate the final compressed output file.

[0054] The following example illustrates the execution process of this method.

[0055] Suppose there is a fishing boat trajectory containing 1000 trajectory points, with the platform type being "fishing boat", the country being "CN", and the name being "DDG-101". The task changes from "fishing" to "returning home" in the middle of the trajectory.

[0056] Execute S1 to identify and classify the multimodal attributes in the trajectory data.

[0057] Execute S2 to extract 100 key nodes based on preset conditions.

[0058] Execute S3.1, creating a dictionary in the file header, for example: Country: {'CN' -> 1}, Platform Type: {'Purse seine fishing vessel' -> 5}, Task: {'Return voyage' -> 101, 'Fishing' -> 102}. The platform name "DDG-101" is calculated as a hash value.

[0059] For the first critical node, record its node index, the original value of the core spatiotemporal attribute, the encoded values ​​of all low-frequency change attributes (0, 1, 5, Hash('DDG-101'), 102), and the original value of the high-frequency change attribute in sequence. Similarly, when storing other critical nodes, record the node index, the original value of the core spatiotemporal attribute, the low-frequency change attribute, and the high-frequency change attribute.

[0060] Execute steps S3.1, S3.2, and S4. For the remaining 900 non-critical nodes, they are grouped into sets of preceding and subsequent critical nodes. That is, non-critical nodes between any two adjacent critical nodes are grouped together, with these two critical nodes acting as the boundaries of the group. The node with the earlier time setting is the preceding critical node, and the node with the later time setting is the subsequent critical node. For each group of non-critical nodes, when a non-critical node in the group changes its task from "fishing" to "returning," only the attribute index 3 and the dictionary code 101 corresponding to the attribute value are recorded. A sliding window dictionary encoding method is then used to encode this low-frequency change attribute. Simultaneously, for all non-critical nodes, the residual between their position information and the predicted value obtained through linear interpolation based on the previous critical node is calculated, and the residual is quantized into a 16-bit integer.

[0061] Execute S5 to encapsulate the file header, 100 key node data blocks, quantized residual vectors of each group of non-key nodes, and low-frequency change attribute encoding data in a hierarchical structure, and perform entropy encoding to generate the final compressed file.

[0062] The decompression process is the reverse of the encoding process. First, the file header is decoded to reconstruct the dictionary and initial state. The data body is read sequentially. When a critical node data block is encountered, its state is completely reconstructed by parsing its full information. When a differential data block is encountered, the positions of all non-critical nodes in the group are interpolated using the core spatiotemporal attributes of the previous critical node and the residual vector in the differential data block. At the same time, combined with differential information and low-frequency change attribute records, the heading, speed, and other attribute information of all non-critical nodes are completely recovered. The heading and speed can also be calculated based on the position and timestamp. The entire process is semantically lossless.

Claims

1. A trajectory compression method for multimodal attributes of moving targets in the air and sea, characterized in that, include: Receive the raw trajectory data stream, parse and extract the multimodal attributes, and classify the multimodal attributes into core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes; Based on preset conditions, key nodes are extracted from the original trajectory data stream, and the node index, core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes of the key nodes are recorded. The low-frequency change attributes are then encoded. For non-critical nodes between any adjacent critical nodes, only record and encode the low-frequency change attributes of the non-critical nodes. For the location information in the core spatiotemporal attributes of non-critical nodes, calculate the residual between it and the predicted value based on the location information obtained from the previous critical node, and quantify the residual. The data recorded and processed by critical and non-critical nodes are encapsulated in a hierarchical structure, and the final compressed output is generated by applying an entropy coding algorithm.

2. The method according to claim 1, characterized in that: The core spatiotemporal attributes include timestamp, longitude, latitude, and altitude; low-frequency change attributes include country, platform name, platform type, and mission; high-frequency change attributes include speed and heading.

3. The method according to claim 1, characterized in that: The preset conditions for extracting key nodes are: the spatial Euclidean distance, rate of change of velocity, and heading deflection angle of the previous key node all exceed the corresponding preset thresholds; the preset thresholds used to filter key nodes are adjusted according to the task type.

4. The method according to claim 1, characterized in that: A global dictionary table is created for low-frequency change attributes, and all low-frequency change data of critical nodes are initially encoded based on this table. For non-critical nodes, the low-frequency change attribute values ​​are only initially encoded using the global dictionary table when they change compared to the preceding critical nodes. After the initial encoding is completed, the sliding window dictionary encoding method is used to encode all the initial encoded values ​​a second time.

5. The method according to claim 1, characterized in that: For non-critical nodes, the predicted value of the location information is obtained by linear interpolation based on the timestamp of the previous critical node and the non-critical node. The residual is obtained by calculating the difference between the actual location information and the predicted value of the non-critical node, and then the residual is quantified.

6. The method according to claim 1, characterized in that, The operation of quantizing the residual is as follows: the calculated floating-point residual value is quantized into an integer with a preset number of bits.

7. The method according to claim 1, characterized in that: The hierarchical structure includes a file header and a data body. The file header includes a dictionary table created for low-frequency change attributes and complete information about the initial key nodes.

8. The method according to claim 7, characterized in that: The data body stores key node data blocks and differential data blocks alternately in chronological order. The differential data blocks contain quantized residual vectors of non-key nodes between adjacent key nodes and encoded data of low-frequency change attributes.

9. A trajectory compression system for multimodal attributes of moving targets in the air and sea, characterized in that, include: Data preprocessing module: used to receive raw trajectory data stream, parse and extract multimodal attributes, and classify multimodal attributes into core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes; Key node analysis module: used to extract key nodes from the original trajectory data stream based on preset conditions, and record the node index, core spatiotemporal attributes, low-frequency change attributes and high-frequency change attributes of the key nodes; Attribute compression module: used to encode low-frequency change attributes of key nodes; For non-critical nodes between any adjacent critical nodes, only the low-frequency change attributes of the non-critical nodes are encoded. For the location information in the core spatiotemporal attributes of non-critical nodes, calculate the residual between it and the predicted value based on the location information obtained from the previous critical node, and quantify the residual. Data encapsulation module: This module encapsulates the data recorded and processed by key nodes and non-key nodes in a hierarchical structure, and applies an entropy coding algorithm to generate the final compressed output.

10. A computer-readable storage medium for storing one or more programs, characterized in that: The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform the method according to any one of claims 1 to 8.

Images