Airport video data real-time analysis system

Abstract

The invention relates to the technical field of airport safety monitoring, and discloses an airport video data real-time analysis system. The system comprises a video stream spatial-temporal feature modeling module, a behavior trajectory map construction module, an abnormal region association analysis module, a risk level semantic judgment module and a situation structure visualization module. According to the method, multi-scale spatial-temporal feature analysis is carried out on an airport monitoring video stream, a multi-dimensional behavior trajectory map is established, abnormal behavior region association is analyzed, risk level semantics are judged, and finally an airport global risk situation thermodynamic distribution map is generated. According to the system, the whole process processing from video data acquisition to risk situation visualization is realized, the abnormal behavior area can be accurately identified, the risk level and category are clear, comprehensive and visual situation information is provided for airport safety management, and the intelligent level of airport safety management is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of airport safety monitoring, in particular to an airport video data real-time analysis system. BACKGROUND

[0002] With the rapid development of the aviation transportation industry, airports, as a personnel-intensive and highly mobile transportation hub, face unprecedented challenges in safety management. Traditional airport video monitoring systems rely on manual inspection or single-dimensional behavior recognition technology, making it difficult to meet the needs of multi-target dynamic tracking and global risk warning in complex scenarios. Currently, most monitoring and analysis schemes in the prior art can only achieve local behavior detection in a single scene, such as identifying single events such as personnel gathering and object leaving in a specific area, and lack of correlation analysis between different monitoring areas, making it difficult to capture abnormal behavior across regions in a timely manner. In practical applications, an airport monitoring network is usually composed of hundreds or even thousands of monitoring cameras. The monitoring angles of different cameras differ, and the monitoring ranges are independent of each other, forming multiple information islands. When abnormal behavior involves multiple monitoring areas, existing systems cannot effectively integrate cross-angle behavior trajectory data, making it difficult to build a global behavior graph from a macro to a micro perspective. At the same time, traditional abnormal detection methods are mostly based on preset threshold judgments, such as setting fixed parameters such as personnel stay time and moving speed to identify abnormalities. This approach has poor adaptability to complex and variable behavior patterns, and is prone to missed or false detections. The prior art lacks semantic description ability in risk level determination, and can only output simple alarm signals, which cannot provide detailed risk categories, levels, and associated areas for management personnel, resulting in low emergency response efficiency. In terms of situation visualization, existing systems mostly use single video frames or simple charts for display, which cannot intuitively present the global risk distribution of the airport, and are not conducive to managers to fully grasp the security situation and make accurate decisions. SUMMARY

[0003] The purpose of the present application is to provide an airport video data real-time analysis system to solve the problems raised in the background art.

[0004] To achieve the above purpose, the present application provides an airport video data real-time analysis system, which comprises: A video stream spatiotemporal feature modeling module is used to receive airport monitoring video stream data, analyze multi-scale spatiotemporal features of continuous video frames, extract motion target trajectory segments and background static region optical flow distribution, identify the occurrence time sequence and spatial coordinates of key behavior events, and generate a spatiotemporal behavior feature sequence. a behavior trajectory atlas construction module, based on the spatiotemporal behavior feature sequence, performing hierarchical behavior modeling on target motion trajectory and regional optical flow change, establishing a topological connection path from a macro region to a micro target, collecting spatial correlation strength and time continuity attributes between path nodes, and outputting a multi-dimensional behavior trajectory atlas structure; an abnormal region association analysis module, according to the multi-dimensional behavior trajectory atlas structure, extracting a behavior path node set under different monitoring perspectives, calculating a behavior mode coincidence degree of cross-region nodes, screening end region coordinates with high frequency, and forming an abnormal behavior region association set; a risk level semantic judgment module, based on the end region coordinates in the abnormal behavior region association set, obtaining predefined risk semantic labels of corresponding regions, prioritizing according to region risk association frequency, matching risk categories of behavior trajectory end region, and generating a multi-level risk semantic label group; a situation structure visualization module, according to the multi-level risk semantic label group, mapping the physical monitoring region nodes corresponding to each label, dividing the behavior trajectory data into region nodes according to risk level, constructing the spatiotemporal mapping relationship between region nodes and behavior trajectories, and generating an airport global risk situation heat distribution map.

[0005] Preferably, the spatiotemporal behavior feature sequence includes a motion target trajectory segment set, a regional optical flow intensity distribution matrix, and a behavior event spatiotemporal coordinate index, the multi-dimensional behavior trajectory atlas structure includes a hierarchical behavior node mapping relationship, a spatiotemporal topological connection chain, and a node association strength parameter set, the abnormal behavior region association set includes a cross-region coincident node set, an end region coordinate frequency distribution, and an associated region screening result, the multi-level risk semantic label group includes a risk label frequency ordering table, an end region risk label mapping relationship, and a risk level matching result, and the airport global risk situation heat distribution map includes a region node spatial identifier, a behavior trajectory spatiotemporal grouping result, and a risk situation mapping structure.

[0006] Preferably, the video stream spatiotemporal feature modeling module includes: a multi-scale feature analysis sub-module, which acquires airport monitoring video stream data, uses a three-dimensional convolution network to decompose the spatiotemporal features of continuous video frames, separates motion target trajectory vectors and background optical flow field distribution matrices, records the starting frame number and spatial bounding box coordinates of each behavior event on the time axis, calculates the proportional relationship between the first trigger position and the number of continuous frames of key behaviors in the spatiotemporal sequence, and generates a behavior event spatiotemporal distribution index; a dynamic segment reconstruction sub-module, which, based on the behavior event spatiotemporal distribution index, intercepts video frame sequences before and after the trigger of key behaviors, performs spatiotemporal alignment operations on motion trajectory vectors and optical flow matrices according to behavior types, integrates multi-event segments under the same behavior category to form a spatiotemporal feature segment set; The feature sequence generation submodule is configured to: count a triggering frequency of each type of behavior event in a unit time window according to the set of spatiotemporal feature segments; rearrange the set of feature segments in a behavior triggering time sequence; splice multiple event segments of the same type according to a first triggering position into a continuous sequence; calibrate a time offset of the behavior feature sequence by using a dynamic time warping algorithm; and output a standardized spatiotemporal behavior feature sequence.

[0007] Preferably, the behavior trajectory graph construction module comprises: The behavior level mapping submodule is configured to: load a predefined airport area spatial topology weight based on the standardized spatiotemporal behavior feature sequence; perform level priority sorting on the motion trajectory nodes and the optical flow area nodes; and establish a node rearrangement index from a top area of the terminal to a final area of the boarding gate. The topology path generation submodule is configured to: extract a spatiotemporal correlation strength parameter between adjacent level nodes according to the node rearrangement index; record a bidirectional connection weight and a time sequence continuity index for each node; calculate a spatial tightness and a time sequence consistency coefficient of a node path by using a graph embedding algorithm; and generate a spatiotemporal topology path with a weight attribute. The graph structure extraction submodule is configured to: collect spatial coordinate connection relationships and time continuity parameters of all nodes in the path based on the spatiotemporal topology path with the weight attribute; and construct a multi-dimensional behavior trajectory graph structure comprising a node spatial position mapping table and a time correlation matrix.

[0008] Preferably, the abnormal area correlation analysis module comprises: The multi-view node collection submodule is configured to: extract a set of behavior path terminal nodes of different camera coverage areas based on the multi-dimensional behavior trajectory graph structure; and record a monitoring view number corresponding to each terminal node and a path duration parameter. The cross-area overlap analysis submodule is configured to: call node spatial coordinate sequences of any two monitoring views according to the set of terminal nodes; calculate a distribution overlap degree of the nodes in a spatial grid; extract end coordinate points that frequently appear in an overlap area; count a difference in occurrence frequency of the coordinate points in multi-view paths; and filter an abnormal area coordinate set with a spatial aggregation exceeding a preset threshold by using a spatiotemporal density clustering algorithm. The correlation set generation submodule is configured to: backtrack a node number of a corresponding coordinate in an original behavior path based on the abnormal area coordinate set; integrate a spatial mapping relationship of a monitoring view number and an area coordinate; and output an abnormal behavior area correlation set comprising an area coordinate frequency distribution graph.

[0009] Preferably, the risk level semantic judgment module comprises: A regional semantic mapping submodule queries a risk semantic label library predefined in an airport electronic fence database based on end point region coordinates in the abnormal behavior region association set, and establishes an index mapping table of region coordinates and risk labels; A risk frequency statistics submodule, according to the index mapping table, counts the association frequency values of each type of risk label in the abnormal region coordinate set, and generates a risk label priority sequence in descending order of frequency values; A level matching submodule matches each abnormal region coordinate with the risk label category with the highest association frequency based on the risk label priority sequence, integrates the label matching results of all region coordinates, and generates a multi-level risk semantic label group containing risk level weight values.

[0010] Preferably, the situation structure visualization module comprises: A physical region mapping submodule analyzes the airport physical region node numbers corresponding to each risk label based on the multi-level risk semantic label group, and records the spatial coordinate range of the node in the electronic map and the number of existing risk trajectories; A trajectory space division submodule groups and maps behavior trajectory spatio-temporal data to corresponding geographical regions according to the airport physical region node numbers, and establishes a bidirectional spatial index structure of trajectory data and geographical regions; A heat map generation submodule calculates the risk trajectory density value in each geographical region based on the bidirectional spatial index structure, maps the density value to a color gradient parameter, and superimposes the airport electronic map to generate a visual heat map layer containing real-time risk intensity distribution.

[0011] Preferably, the system further comprises a real-time early warning feedback module, which, based on the airport global risk situation heat distribution map, performs the following operations: Extracts a set of physical region coordinates in the heat map whose risk intensity exceeds a preset threshold; Determines the risk type priority according to the multi-level risk semantic label group associated with the set of physical region coordinates; Generates an early warning instruction data packet containing risk coordinate position, risk type, and suggested disposal measures; Pushes the early warning instruction data packet to the corresponding regional terminal device in real time.

[0012] Preferably, the real-time early warning feedback module further comprises: An early warning verification submodule receives disposal result data fed back by the terminal; Compares the original risk intensity value of the region to be verified with the real-time risk intensity value after disposal; When the risk intensity reduction rate does not reach a preset target, automatically upgrades the early warning level and triggers a secondary disposal instruction.

[0013] Preferably, the system further comprises an adaptive feature updating module for performing: periodically collecting the regional risk intensity change rate in the airport global risk situation heat distribution map; triggering the video stream feature re-extraction instruction of the corresponding region when the risk intensity change rate of a specific physical region continuously exceeds the dynamic threshold value; reconstructing the node correlation strength parameter in the multi-dimensional behavior trajectory graph structure based on the re-extracted spatio-temporal behavior feature sequence; updating the cross-region behavior pattern coincidence degree calculation model in the abnormal region correlation analysis module according to the reconstructed node correlation strength parameter; outputting the updated behavior pattern coincidence degree calculation model to the risk level semantic judgment module, and synchronously optimizing the generation logic of the risk label priority sequence.

[0014] Compared with the prior art, the beneficial effects of the present application are: Through the multi-scale spatio-temporal feature analysis of the continuous video frames by the video stream spatio-temporal feature modeling module, the moving target trajectory segment and the background static region optical flow distribution can be accurately extracted, which provides detailed and comprehensive basic data for subsequent behavior analysis, making the identification of key behavior events more accurate, and helping to capture subtle behavior changes that are easily overlooked in traditional monitoring. The behavior trajectory graph construction module performs hierarchical behavior modeling based on the spatio-temporal behavior feature sequence, and establishes a topological connection path from a macro region to a micro target. This hierarchical modeling method can associate the dispersed target motion trajectory with the region optical flow change, forming a complete behavior trajectory graph structure, so that the management personnel can clearly understand the movement path of different targets in the airport and their association, and can grasp the behavior dynamics in the airport as a whole. The abnormal region correlation analysis module extracts the cross-region behavior path node set and calculates the coincidence degree according to the multi-dimensional behavior trajectory graph structure, and screens the high-frequency end region, which can effectively find abnormal behavior regions that are associated under different monitoring perspectives, breaking the situation of isolated information in each region in traditional monitoring, making the cross-region abnormal behavior tracking more efficient, and helping to timely find potential security risk areas. The risk level semantic judgment module obtains the pre-defined risk semantic labels and performs priority sorting based on the abnormal behavior region association set, matches the risk categories to generate a multi-level label group, so that the risk judgment is no longer a simple alarm, but can clearly determine the category and level of the risk, so that the management personnel can take corresponding measures according to different risk levels and categories, enhancing the pertinence and effectiveness of risk handling. The situation structure visualization module maps the multi-level risk semantic label group to the physical monitoring area node, constructs the space-time mapping relationship and generates a global risk situation heat distribution map. This kind of visual presentation can intuitively display complex risk information, so that the management personnel can quickly master the risk distribution of the whole airport, understand the risk level of different areas and the associated behavior trajectory, so as to more efficiently allocate resources and improve the overall efficiency of airport safety management. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A working principle diagram of the airport video data real-time analysis system is provided. Figure 2 A flowchart of the video stream space-time feature modeling module is provided. Figure 3 A flowchart of the behavior trajectory graph construction module is provided. Figure 4 A flowchart of the risk level semantic determination module is provided. DETAILED DESCRIPTION The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0016] Please refer to Figure 1 The present application provides an airport video data real-time analysis system, which comprises: The video stream space-time feature modeling module receives the airport monitoring video stream data, analyzes the multi-scale space-time features of the continuous video frames, extracts the motion target trajectory segments and the background static area optical flow distribution, identifies the occurrence time sequence and spatial coordinates of the key behavior events, and generates the space-time behavior feature sequence.

[0017] The behavior trajectory graph construction module performs hierarchical behavior modeling on the target motion trajectory and the area optical flow change based on the sequence, establishes the topological connection path from the macro area to the micro target, collects the spatial correlation strength and time continuity attributes between the path nodes, and outputs the multi-dimensional behavior trajectory graph structure.

[0018] The abnormal area association analysis module extracts the behavior path node set under different monitoring angles according to the graph structure, calculates the behavior mode coincidence degree of the cross-area nodes, filters the high-frequency end point area coordinates, and forms the abnormal behavior area association set.

[0019] The risk level semantic judgment module obtains a predefined risk semantic label corresponding to the region based on the end region coordinates in the set, prioritizes the risk association frequency of the region, matches the risk category of the behavior trajectory end region, and generates a multi-level risk semantic label group.

[0020] The situation structure visualization module maps the physical monitoring region nodes corresponding to each label according to the label group, divides the behavior trajectory data into the region nodes according to the risk level, constructs the spatio-temporal mapping relationship between the region nodes and the behavior trajectory, and generates an airport global risk situation heat distribution map.

[0021] Embodiment 1: refer to Figure 2 The video stream spatio-temporal feature modeling module of the airport video data real-time analysis system is responsible for multi-scale spatio-temporal feature analysis of the input monitoring video stream, and generates a standardized spatio-temporal behavior feature sequence. The module is composed of a multi-scale feature analysis submodule, a dynamic segment reconstruction submodule and a feature sequence generation submodule, and each submodule works cooperatively to complete the conversion from the original video data to the structured feature sequence.

[0022] The multi-scale feature analysis submodule receives real-time video stream data transmitted by the airport monitoring camera, and uses a three-dimensional convolution network to perform spatio-temporal feature decomposition on continuous video frames. The network analyzes the inter-frame motion changes in the time dimension and extracts the target contour and background optical flow distribution in the space dimension. The trajectory information of the moving target is quantized into a vector form, recording the displacement direction and speed change of the target in the continuous frames. The background optical flow field distribution is stored in a matrix form, reflecting the macro motion trend of the static region. Each detected behavior event is marked with the starting frame number and the spatial bounding box coordinates, and the bounding box is used to locate the physical region where the behavior occurs. The spatio-temporal distribution of the key behavior is quantified by calculating the proportional relationship between the first triggering position and the number of continuous frames, generating a behavior event spatio-temporal distribution index. The index takes the time axis as the reference, recording the triggering time sequence and spatial coverage range of different behavior categories.

[0023] The dynamic segment reconstruction submodule extracts the video frame sequence within a specific time window before and after the triggering of the key behavior according to the behavior event spatio-temporal distribution index. The extracted segment needs to contain the complete period of the behavior, and the entire process from the initial triggering to the end is retained. The trajectory vector of the moving target and the background optical flow matrix perform alignment operation in the spatio-temporal dimension, eliminating the spatial deviation caused by different camera perspectives. Multiple event segments of the same behavior category are integrated into a unified feature representation, forming a spatio-temporal feature segment set. The set is stored by behavior type, and the segments under each category have consistent spatio-temporal feature dimensions.

[0024] The feature sequence generation submodule statistically analyzes the spatio-temporal feature fragment set and calculates the triggering frequency of each type of behavior event in a unit time window. The behavior events are rearranged in chronological order to ensure continuity on the time axis. Multiple event fragments of the same category are spliced according to the first triggering position to form a continuous behavior feature sequence. The dynamic time warping algorithm is used to calibrate the timing offset between different instances and eliminate the time difference caused by detection delay or video stream jitter. The calibrated feature sequence has a unified time reference, and the standardized spatio-temporal behavior feature sequence is finally output.

[0025] The spatio-temporal behavior feature sequence includes three main components. The set of moving target trajectory fragments records the motion paths of all detected targets and is stored in the form of a vector sequence, each vector containing the position and motion state of the target in a single frame. The regional optical flow intensity distribution matrix stores the macroscopic motion pattern of the background region. The rows and columns of the matrix correspond to the spatial grid division of the video frames, and the matrix element values represent the optical flow intensity of the corresponding grid. The behavior event spatio-temporal coordinate index adopts a tree structure organization, with the root node recording the behavior category and the child nodes storing the starting frame number and bounding box coordinates of all instances under this category.

[0026] The multi-dimensional behavior trajectory graph structure is composed of hierarchical behavior node mapping relationship, spatio-temporal topology connection chain, and node association strength parameter set. The hierarchical behavior node mapping relationship adopts a tree structure organization, with the top-level nodes corresponding to the airport functional areas, the middle-level nodes representing specific functional partitions, and the bottom-level nodes associated with specific monitoring points. The spatio-temporal topology connection chain records the transfer path between nodes, with each chain storing the source node, target node, transfer time, and spatial distance. The node association strength parameter set includes the spatial tightness coefficient and the time continuity index, quantifying the degree of association between nodes.

[0027] The abnormal behavior region association set is generated through cross-region node analysis. The cross-region overlapping node set stores the same physical location nodes detected under different camera perspectives, with each node labeled with the camera number and appearance time. The end region coordinate frequency distribution statistics all behavior path termination positions, recording the number of occurrences of each position in different time periods. The associated region screening result is generated using a spatial clustering algorithm, labeling the abnormal region coordinates and associated behavior categories with high frequency aggregation.

[0028] The multi-level risk semantic label group contains three types of structured data. The risk label frequency ranking table ranks all detected risk types in descending order of frequency, with each type labeled with its triggering frequency within the analysis period. The end region risk label mapping relationship is stored in a hash table, with the key being the region coordinate and the value being all risk labels associated with the coordinate and their occurrence frequencies. The risk level matching result records the final determined risk category of each abnormal region, including the risk level weight value and the determination basis.

[0029] The airport-wide risk situation thermodynamic distribution map is composed of three parts of data. The regional node space identification adopts geographic information system coordinates, accurately marking the physical boundaries of each monitoring area. The behavior trajectory space-time grouping results are stored by risk level classification, and each group of trajectory data is labeled with the risk category and the time period. The risk situation mapping structure is stored in a graph database, with nodes representing physical areas and edges representing risk transmission paths, and edge weights reflecting risk transmission intensity.

[0030] The operation of the video stream space-time feature modeling module does not rely on manual intervention, and all processing steps are automatically executed. The parameters of the three-dimensional convolutional network are obtained by pre-training with historical monitoring data, which can adapt to the environmental differences of different airports. The behavior event detection algorithm supports online updating, and when new behavior categories are added, new detection models can be dynamically loaded. The space-time alignment operation is realized by using the feature point matching algorithm, which is robust to changes in camera perspective. The sliding window size of the dynamic time warping algorithm is adaptively adjusted according to the video frame rate to ensure the timing calibration accuracy under different sampling rates. The output format of the standardized feature sequence is compatible with the subsequent modules and can be directly used for behavior trajectory graph construction.

[0031] In the multi-scale feature analysis process, the moving target detection uses a lightweight neural network to ensure accuracy while meeting real-time requirements. The background optical flow calculation uses a sparse optical flow algorithm, focusing on monitoring the light variation patterns in densely populated areas. The behavior event space-time distribution index uses time encoding technology for compressed storage, reducing memory usage. The behavior classifier in the dynamic segment reconstruction stage supports incremental learning, allowing the behavior category system to be gradually improved. The frequency statistics during feature sequence generation use a sliding time window, and the window size is dynamically adjusted according to the airport operating status.

[0032] The storage of space-time behavior feature sequences uses a columnar database to optimize the read-write efficiency of large-scale trajectory data. The regional optical flow matrix uses sparse matrix compression technology, storing only non-zero elements and their coordinates. The behavior event index tree realizes fast range queries, supporting event retrieval by time and space conditions. The construction process of multi-dimensional behavior trajectory graphs uses a distributed computing framework, supporting parallel processing of massive nodes and edges. The abnormal area detection algorithm introduces spatial topology constraints to avoid misjudging normal transfer areas as abnormal. The risk level matching integrates multiple sources of data, including historical event records and real-time sensor readings. The generation of the thermodynamic distribution map uses GPU accelerated rendering to ensure real-time visualization of large-scale scenes.

[0033] The hardware implementation of this module is based on a heterogeneous computing architecture, with the CPU handling logical control and data scheduling, and the GPU accelerating deep learning inference and image processing. An in-memory database is used to cache intermediate results, reducing disk IO latency. Network communication uses zero-copy technology to optimize the transmission efficiency of video stream data. The internal sub-modules of the module communicate asynchronously through message queues, achieving loose coupling and high concurrency processing. The error handling mechanism monitors the running state of each link and automatically starts the recovery process in the event of an abnormal situation. The performance monitoring system collects processing delays and resource utilization in real time, providing data support for system optimization.

[0034] The design of the video stream spatio-temporal feature modeling module takes into account the special needs of airport monitoring scenarios. Changes in lighting conditions are automatically corrected through a white balance algorithm, avoiding the impact of day-night transitions on feature extraction. The problem of target occlusion in crowded scenes is alleviated through a multi-target tracking algorithm, maintaining trajectory continuity. Camera jitter interference is eliminated using electronic image stabilization technology, ensuring the accuracy of motion detection. The compatibility of video sources with different resolutions is achieved through an adaptive scaling mechanism, allowing for the unified processing of high-definition and standard-definition video streams. The operational characteristics of airports in different time zones are addressed through the use of UTC time standardization, unifying the time base of all data.

[0035] Example 2: Referring to Figure 3 The behavior trajectory atlas construction module of the airport video data real-time analysis system works in conjunction with the abnormal region association analysis module to convert standardized spatio-temporal behavior feature sequences into structured behavior trajectory atlases and identify abnormal region association sets from them. These two modules achieve accurate portrayal of personnel flow patterns and potential risk areas within the airport through hierarchical modeling and cross-perspective analysis.

[0036] The behavior trajectory atlas construction module first processes the input spatio-temporal behavior feature sequences through the behavior level mapping sub-module. This sub-module loads predefined airport spatial topology weight data, which reflect the importance and connectivity of each functional area within the terminal. Trajectory nodes of moving targets are classified according to their physical regions, while background optical flow region nodes are divided based on the monitoring field of view. The node level system is constructed from macro to micro, with top-level nodes corresponding to terminal overall partitions such as departure hall, security area, and boarding area. Middle-level nodes are further refined to specific functional areas such as check-in counter and baggage carousel. Bottom-level nodes are precise to micro regions covered by individual cameras. Each node is assigned a unique level index, forming a complete spatial topology structure.

[0037] The topology path generation submodule extracts the transition relationship between adjacent hierarchical nodes based on hierarchical index expansion analysis. For each pair of nodes with spatio-temporal correlation, the bidirectional connection weight is recorded, which integrates factors such as spatial distance, transition frequency, and time interval. The time sequence continuity index is calculated by analyzing the time regularity of node transition in historical data, reflecting the stability of the path. The graph embedding algorithm projects the nodes into a low-dimensional vector space, capturing the time evolution pattern while preserving spatial proximity. The generated vectors are used to calculate the tightness score of the node path. The spatio-temporal topology path is finally stored in the form of a weighted directed graph, with edge weights including spatial correlation strength and time continuity coefficient.

[0038] The atlas structure extraction submodule integrates the generated topology path to construct a multi-dimensional behavior trajectory atlas. The node spatial position mapping table records the precise coordinates of all nodes in the airport electronic map and their respective hierarchical levels. The time correlation matrix stores the time sequence relationship between nodes in sparse matrix format, with matrix element values representing the statistical characteristics of the transition time interval. The complete atlas structure supports multi-dimensional queries, allowing both macro-level analysis of regional passenger flow trends and micro-level tracking of specific target movement paths. The atlas update mechanism uses incremental processing, with newly arrived spatio-temporal behavior features being integrated into the existing structure in real time.

[0039] The abnormal area correlation analysis module works based on the multi-dimensional behavior trajectory atlas. The multi-view node collection submodule first extracts the behavior path terminal nodes from different camera perspectives, which represent the final resting positions of the target within the monitoring field. Each terminal node is associated with its corresponding camera number, arrival timestamp, and duration of stay. The node data is normalized to eliminate the influence of different camera time synchronization errors. The cross-region overlap analysis submodule performs spatial matching on the multi-view terminal nodes, converting the coordinate systems of each camera to the global coordinate system of the airport.

[0040] The spatial grid division strategy divides the airport plane into uniform cells, and counts the frequency of nodes from different perspectives appearing in each cell. The distribution overlap algorithm calculates the similarity of node distribution in the same grid for any two perspectives, identifying high-frequency areas of common interest from multiple perspectives. The spatio-temporal density clustering algorithm performs secondary analysis on these areas, considering the time distribution characteristics of node appearance, and selects coordinate points that exhibit clustering patterns in both time and space. The persistence verification is introduced during the generation of the abnormal area coordinate set, and only those areas that repeatedly appear within consecutive time windows are retained.

[0041] The associated set generation submodule performs backtracking analysis on the filtered abnormal regions to find their corresponding nodes in the original behavior trajectory graph. Each abnormal region is associated with complete information of its source path, including the sequence of nodes passed through, the sequence of transfer times, and the list of associated cameras. The region coordinate frequency distribution graph is visualized in the form of a heat map, which intuitively shows the abnormality degree of different regions. The output of the abnormal behavior region association set uses a hierarchical storage structure, with the top layer recording the spatial distribution characteristics of the region, the middle layer storing the path backtracking information, and the bottom layer retaining the original node data.

[0042] The behavior trajectory graph construction module is implemented using a distributed graph computing framework, supporting parallel processing of large-scale nodes and edges. The hierarchical index construction process introduces spatial indexing to accelerate region queries, using an R-tree structure to optimize query efficiency. The graph embedding training process uses negative sampling techniques to reduce computational complexity without sacrificing accuracy. The weight calculation of the spatio-temporal topology path uses a sliding time window to calculate the window size dynamically according to the airport operating period. The incremental update mechanism is realized through a difference detection algorithm, which only recalculates the local graph affected by new data.

[0043] The multi-perspective data processing of the abnormal region association analysis module uses a time alignment algorithm to compensate for the time deviation caused by network transmission delay. The granularity of the spatial grid division is automatically adjusted according to the airport passenger flow, with finer grids used during peak periods to improve positioning accuracy. The parameters of the density clustering algorithm are learned from historical data to adapt to the spatial distribution characteristics of different regions. The abnormal region verification stage introduces behavior pattern comparison to analyze the similarity between the current clustering pattern and historical normal patterns. The storage of the associated set uses columnar compression format to optimize the access efficiency of high-frequency coordinate points.

[0044] Error handling and recovery mechanism is improved. Input data verification uses pattern matching, illegal format is immediately isolated processing. The key state of the processing process is periodically persisted, and it is recovered from the latest consistent state after failure. Resource overrun triggers automatic degradation, prioritizing core function operation. Network partition detection is realized through the heartbeat mechanism, and the local cache mode is switched when abnormal. Security audit logs record all data access tracks, supporting post-trace analysis.

[0045] The scalability design of the module supports future demand changes. The behavior graph model reserves attribute extension interfaces, which can add dimensions without affecting existing logic. The analysis algorithm is designed as a plug-in, supporting dynamic loading of different implementations. The storage backend is abstracted as a unified interface, which can adapt to different database technologies. The computing resource pool supports horizontal expansion, adding processing nodes as needed. Protocol version compatibility is guaranteed through an adaptation layer, and new and old systems can coexist and run.

[0046] Example 3: see Figure 4The risk level semantic judgment module and the situation structure visualization module of the airport video data real-time analysis system work together to transform the set of abnormal behavior area associations into multi-level risk semantic label groups, and finally generate a heat map of the risk situation distribution across the entire airport. These two modules, through semantic mapping and spatial visualization technologies, achieve a quantitative assessment and intuitive presentation of airport security risks.

[0047] The initial processing of the risk level semantic determination module is completed by the regional semantic mapping submodule. This submodule connects to the airport's electronic fence database, which stores predefined risk attributes for all functional areas of the terminal. Each physical area is recorded in the database with a unique geocode, associated with three types of basic information: area function type (e.g., security checkpoint, baggage claim area), historical risk event statistics, and preset risk semantic tags. The tagging system adopts a hierarchical classification method, with the top level consisting of three risk levels: major, medium, and general. The lower level is further subdivided into 12 specific risk categories, including crowd gathering, abnormal detention, and reverse flow. The mapping process uses a bidirectional index structure; the forward index quickly finds tags through geocodes, while the reverse index supports searching related areas by tag.

[0048] The risk frequency statistics submodule performs multi-dimensional analysis on the coordinate set of abnormal areas. Each coordinate point carries timestamp information, and a time decay factor is introduced during statistics, with recent events having a higher weight than historical events. Frequency statistics adopt a sliding time window mechanism, with the window width dynamically adjusted according to the airport's operational status. The window is set to 30 minutes for normal periods and automatically shortened to 15 minutes for special periods (such as large-scale flight delays). During the generation of the risk label priority sequence, conflicts are resolved for coordinate points simultaneously associated with multiple labels, prioritizing labels with higher relevance to the area's function. The frequency statistics results are normalized using the following formula: in, Indicates risk label The normalized frequency probability, For the region The spatiotemporal weighting coefficients, It is an indicator function (when) and The value is 1 when associated, otherwise 0. This represents the total number of abnormal regions. Total number of risk label types. Spatiotemporal weighting coefficient. It is derived by weighting three factors: area, population density, and event duration.

[0049] The level matching sub-module performs the final risk judgment. Each abnormal area coordinate matches its top three risk labels with the highest correlation to form a candidate label set. The candidate labels are filtered by a rule engine to eliminate options that are obviously inconsistent with the area's function (e.g., a baggage claim area will not have a "check-in queue over limit" label). The matching results are attached with a confidence score, which combines the frequency probability, area function compatibility, and historical verification accuracy. The output of the multi-level risk semantic label group is organized in a tree structure, with the root node being the risk level, the child nodes being specific risk types, and the leaf nodes storing the associated area coordinate list and matching parameters.

[0050] The physical area mapping sub-module of the situation structure visualization module analyzes the multi-level risk semantic label group and converts abstract labels into specific geographic elements. The airport electronic map uses a hierarchical vector data model, including building contour layer, facility point layer, and traffic path layer, among other basic geographic information. Each risk label is associated with a specific map element type, such as the "person gathering" label mapping to an open area polygon and the "equipment anomaly" label associated with a facility point. The area node numbering system is consistent with the airport asset management system, ensuring the accuracy of cross-system data correlation. The number of existing risk trajectories is obtained through a spatial connection query to count the amount of historical trajectory data within each geographic element range.

[0051] The trajectory space division sub-module establishes a bidirectional index between behavior trajectories and geographic areas. Risk level weights are used as the basis for division, dividing trajectory data into three priority queues. The spatial index uses an improved R* tree structure, adding a risk level dimension to the traditional spatial division to form a three-dimensional index space. The construction process of the bidirectional index uses a batch loading strategy, first dividing the space into coarse blocks based on spatial range, and then fine-tuning within each block based on risk level. The index update mechanism supports incremental maintenance, with newly arrived trajectory data triggering local index reconstruction to avoid the overhead of global reconstruction.

[0052] The core of the heat map generation sub-module is the calculation of the risk density field. Geographic areas are discretized into 50cm x 50cm grid cells, and the risk density value of each cell is composed of three components: real-time trajectory density , historical baseline density , and risk level weight . The synthesis calculation uses a non-linear superposition model to highlight the visual contrast of high-risk areas. Color mapping uses the HSL color space, with hue representing risk type (red for personnel-related risks and blue for equipment-related risks), saturation reflecting density absolute value, and brightness adjusting automatically based on ambient lighting conditions. The heat map layer is overlaid with the electronic map using Alpha blending technology to preserve the recognizability of the underlying map. The visualization system supports multi-level zooming, dynamically adjusting the rendering granularity and data aggregation level of the heat map at different view levels.

[0053] The hardware deployment of the risk level semantic judgment module adopts a high-availability architecture. The database cluster adopts a master-slave replication mode, and the electronic fence database is deployed on an in-memory computing node to achieve millisecond-level query response. The rule engine runs on a dedicated inference server, and the loaded risk judgment rules support hot updates. The cache system adopts a hierarchical design, with high-frequency access tag mapping tables residing in memory and full data stored on an SSD array. Quality of service control is implemented for computing resource allocation to ensure that the judgment delay during peak periods does not exceed the set threshold.

[0054] The rendering pipeline of the situation structure visualization module is optimized for large-scale geographic data. Vector map data uses GPU instance rendering technology, with tens of thousands of geographic features drawn in a single batch. The heat field calculation utilizes compute shaders for parallel processing, with an independent thread allocated to each grid cell. A dynamic load balancing mechanism monitors the rendering frame rate and automatically reduces the drawing precision of non-critical areas when performance decreases. A multi-view synchronization system maintains consistency in the display content of different terminals, supporting collaborative viewing between the command center large screen and mobile terminals.

[0055] Standardized protocols are used for data interaction between modules. Risk label group transmission uses the Apache Avro binary format, which includes complete schema definitions. The geographic coordinate system uses UTM projection on the WGS84 ellipsoid to ensure spatial analysis accuracy. Time series data is attached with NTP timestamps, with synchronization accuracy controlled within milliseconds. The message queue uses a partitioned mode, with communication channels divided by airport functional areas to reduce unnecessary network transmission.

[0056] Embodiment 4: Real-time warning feedback module of airport video data real-time analysis system Based on the airport global risk situation heat distribution map, the module generates warning instructions for areas exceeding the risk threshold and realizes closed-loop verification of the warning effect. Through a multi-level warning mechanism and a disposal feedback system, the module forms a complete process from risk detection to disposal verification.

[0057] Taking a typical international airport T2 terminal as an example, when the system detects abnormal risk aggregation in the No. 3 gate area of the departure hall, the real-time warning feedback module starts the processing flow. The heat distribution map data shows that the risk intensity value of this area is 87 (the threshold is set to 75), and the associated multi-level risk semantic label group is marked as "abnormal gathering of personnel - secondary risk". The system automatically extracts the physical coordinate information of this area and generates a warning instruction data packet containing the following elements in combination with the spatial encoding of the airport electronic map: The early warning instruction data packet is pushed to three terminals through the airport dedicated communication network: the mobile terminal of the ground service department on-duty supervisor, the large screen system of the terminal central control room, and the emergency broadcast equipment near Gate 3. The data packet is packaged in JSON format, containing standardized field structure and metadata description, ensuring that different systems can be correctly parsed. The mobile terminal triggers a vibration alarm immediately after receiving the data packet, displays the early warning details interface, and automatically retrieves the real-time monitoring screen of the area. The large screen system in the central control room highlights the risk area on the electronic map and synchronously labels the security resource locations within a 500-meter range. The emergency broadcast equipment plays pre-recorded guidance voice, and the volume is automatically adjusted according to the environmental noise level.

[0058] The early warning verification submodule starts the monitoring process after the instruction is issued. The system continuously collects feedback data in three dimensions: the responsibility terminal operation log records the start time and type of the disposal action, the video analysis subsystem monitors the personnel density changes in the target area, and the mobile terminal GPS positioning confirms the arrival of staff on site. If the risk intensity value does not decrease by 15% within 5 minutes after the early warning is issued, the system automatically upgrades the early warning level. Based on the original secondary risk, additional instructions are added to notify the mechanical and electrical engineering department to check the operation status of the gate in the area. The upgraded early warning data packet adds associated elements: historical same period data comparison results, equipment failure possibility analysis, and cross-departmental collaborative disposal guidelines.

[0059] The early warning feedback mechanism includes a multi-level confirmation process. Frontline staff report preliminary disposal through mobile terminals, including on-site description, photo evidence, and preliminary judgment. Professional security personnel submit a second verification report after arriving, detailing the cause of the abnormal gathering (such as queuing caused by temporary malfunction of the check-in system). The central control room attendant marks the event disposal status in the system based on feedback from all parties. All feedback information is stored in a structured manner, forming a complete early warning event disposal archive.

[0060] The abnormal scenario processing mechanism ensures system robustness. When network interruption causes instruction transmission failure, the system automatically switches to a backup communication channel, prioritizing the delivery of early warnings in high-risk areas. In the case of mobile terminal offline, early warning instructions are temporarily stored in the regional edge computing node and will be sent again after connection is restored. For false positives, staff can report false positives through the terminal, and the system records this feedback for optimizing the risk determination model. Regions with multiple false positives will automatically reduce sensitivity and trigger a re-calibration of video analysis parameters.

[0061] The early warning priority dynamic adjustment mechanism optimizes resource allocation according to real-time conditions. When there is a large-scale flight delay, the system temporarily increases the risk monitoring frequency of the check-in area. During special events, the early warning response level of the VIP channel is automatically upgraded. During the night low passenger flow period, the early warning threshold of non-critical areas is appropriately relaxed. The dynamic adjustment parameters are derived from the airport operation database, which synchronizes real-time flight dynamics, special event schedules, and other key information.

[0062] The historical early warning data analysis support system continuously optimizes. Daily automatically generated early warning performance reports contain multiple indicators: average response time, risk resolution success rate, multi-department collaboration efficiency, etc. These data are presented through a dashboard to help managers identify process bottlenecks. The system performs early warning scenario backtracking tests every month to simulate typical risk events and verify the integrity of the disposal process. Test results are used to update the emergency plan knowledge base, supplement new disposal methods and contact information.

[0063] Integration with other airport security systems expands early warning capabilities. When the early warning involves suspicious items, the security X-ray machine historical images are automatically retrieved for comparison. Personnel identity recognition early warning triggers facial recognition system for key monitoring. Fire risk early warning links with the fire control system to open emergency passages in advance. The integrated interface uses the airport's unified middleware platform to ensure secure and reliable data exchange between systems.

[0064] The hardware deployment of the early warning feedback module takes into account actual operation and maintenance needs. The mobile terminal uses industrial-grade drop-resistant design, and the battery life meets the continuous 12-hour work requirement. The central control room server is configured with redundant power supply and network interface to ensure high availability. Edge computing nodes are deployed in each area of the terminal building, achieving localized data processing. All hardware devices are included in the airport asset management system, and regular preventive maintenance is performed.

[0065] The personnel operation interface design conforms to the principles of human engineering. The mobile terminal early warning interface uses red, yellow, and blue color coding to distinguish risk levels, and key information is displayed in large font. The control room large screen system supports gesture zooming operation, and important alerts are automatically popped up to the foreground. The voice broadcast content is acoustically optimized and remains clear and distinguishable in noisy environments. The interface language supports English and Chinese switching to meet the needs of international airport staff.

[0066] The permission management system ensures the safe operation of the early warning system. Different levels of staff are granted different operation permissions: frontline staff can only confirm early warnings and report disposal conditions, supervisors can adjust early warning levels, and system administrators can modify judgment parameters. All operation records are audited in detail and stored with the employee ID. Sensitive operations require secondary authentication, and key parameter modifications require multiple reviews.

[0067] The real-time early warning feedback module shows stable performance characteristics in actual operation. The average delay of a typical early warning from generation to terminal delivery is controlled within 3 seconds, meeting the real-time response requirements. The system supports processing more than 200 early warning events per day, and the concurrent processing during peak periods is buffered through a message queue. The network bandwidth occupancy is optimized, and the average size of the early warning data packet is compressed to below 50KB. The signal coverage rate of the mobile terminal in the entire area of the terminal is 99.7%, ensuring reliable delivery of early warning instructions.

[0068] Example 5: The adaptive feature update module of the airport video data real-time analysis system continuously monitors the risk situation changes, dynamically adjusts the feature extraction and behavior analysis strategy, and ensures that the system adapts to the dynamic changes of the airport operation environment. This module triggers the feature re-extraction and model update process by periodically evaluating the risk intensity change trend, maintaining the accuracy and timeliness of the analysis system.

[0069] After the module is started, a risk intensity baseline is first established. The system loads the historical data of the airport global risk situation heat map for the past 30 days, and calculates the distribution characteristics of the daily risk intensity value for each region. For each functional subarea of the terminal, the average value and standard deviation of the risk intensity are calculated to form the initial setting of the dynamic threshold. The baseline data storage uses a time series database to support efficient range queries and statistical analysis. The risk intensity change rate is monitored with a 15-minute basic time window, and the latest heat value of each region is collected in the window period and compared with the baseline data.

[0070] When the risk intensity change rate of a specific area continuously exceeds the dynamic threshold, the system starts the feature re-extraction process. The change rate evaluation uses a sliding window mechanism, and only when three consecutive monitoring windows exceed the threshold is it considered as a valid trigger to avoid false positives caused by transient fluctuations. The trigger instruction includes a detailed description of the affected area: geographical boundary coordinates, associated camera list, abnormal time period marker. After receiving the instruction, the video stream spatio-temporal feature modeling module prioritizes extracting the original video data of the abnormal period from the cache, and if the cache has expired, it re-extracts from the storage system. The feature extraction process uses incremental update mode, only analyzing the video stream in the abnormal area, and retaining the existing analysis results of other areas.

[0071] The reconstruction process of the spatio-temporal behavior feature sequence introduces a change detection algorithm. The system compares the newly extracted feature sequence with the historical normal pattern to identify trajectories and optical flow distributions that deviate from the norm. For moving target trajectories, analyze their speed change pattern, path deviation degree, and dwell time distribution; for background optical flow fields, detect their intensity distribution changes, direction consistency, and regional correlation. Deviating features are marked as key change points for subsequent atlas structure updates.

[0072] The node correlation strength parameter updating of the multi-dimensional behavior trajectory atlas adopts a gradual adjustment strategy. The system maintains two sets of correlation parameters: long-term stable parameters reflect the inherent connection characteristics between regions, and short-term dynamic parameters capture changes under the current operating state. The updating process first adjusts the short-term dynamic parameters, and when the changes persist for more than a preset period, they are gradually integrated into the long-term parameters. When recalculating the spatio-temporal correlation strength between nodes, factors such as the frequency, duration, and node transfer speed of path appearance during abnormal periods are considered. The graph embedding representation is updated synchronously, fine-tuning the node positions in the low-dimensional vector space to reflect the latest changes in behavior patterns.

[0073] The cross-region behavior pattern overlap calculation model of the abnormal region correlation analysis module is updated accordingly. The system redefines the spatial correlation measure between regions based on the reconstructed node correlation strength. The overlap calculation introduces a time decay factor, with recent behavior patterns having a greater impact on the results than historical data. The model updates use an online learning mechanism, with newly arrived abnormal region data continuously adjusting the classification boundaries to adapt to the evolution of risk patterns. In the calculation process, inherent regional characteristics and temporary changes are distinguished to avoid overfitting to short-term anomalies.

[0074] The risk level semantic judgment module receives the updated behavior pattern information and optimizes its label priority generation logic. The system reevaluates the matching degree of each risk label with the regional characteristics and adjusts the weight distribution of label frequency statistics. The semantic mapping rules are dynamically extended, and when new abnormal behaviors are detected, temporary risk labels are automatically created for manual review. The parameters of the judgment engine are self-adaptively calibrated, maintaining the stability of the overall judgment framework while implementing differentiated judgment strategies for regions with significant changes.

[0075] The resource management of the module adopts an intelligent scheduling strategy. Computing resources are preferentially allocated to update tasks in risk-increasing regions to ensure real-time performance in critical regions. Memory caching dynamically adjusts resident data based on regional change frequency, with feature sequences of high-frequency update regions constantly residing in memory. Network bandwidth allocation implements quality of service control to ensure the transmission stability of video stream re-extraction tasks. The storage system uses a hot and cold data layering mechanism, with data from recently active regions saved in the high-speed storage layer.

[0076] All update operations are recorded with detailed logs, including trigger conditions, processing duration, resource consumption, and other metadata.

[0077] The special scenarios of airport operations are specifically addressed in the module design. During the flight season change period, the sensitivity of change detection is automatically relaxed to adapt to the changes in passenger flow patterns brought about by the new schedule. During major events, the monitoring frequency of VIP channels is temporarily increased to ensure rapid adaptation in key areas. In extreme weather conditions, the judgment threshold of outdoor areas is adjusted to consider the impact of weather on pedestrian behavior. During night maintenance periods, update tasks for non-operational areas are closed to save computing resources.

[0078] The maintainability of the module is improved through standardization. Configuration parameters are managed declaratively, and change history is tracked through version control tools. Monitoring indicators cover various aspects of the update process, including trigger accuracy, feature extraction delay, model convergence state, etc. Diagnostic tools are integrated into the operation and maintenance platform, supporting root cause analysis when updates are abnormal. Automated testing verifies core update logic to ensure that code modifications do not affect basic functions.

[0079] The adaptive feature update module gives the analysis system the ability to evolve dynamically, making it not limited to the limitations of the initial training data. By continuously sensing environmental changes and adjusting itself, the system can adapt to seasonal fluctuations in airport passenger flow patterns, operational strategy adjustments, unexpected events, and other diverse scenarios. This self-updating feature significantly extends the system's effective service period and reduces manual maintenance costs, providing persistent and reliable technical support for airport safety management. The operation of the module forms a closed loop from data to knowledge to optimization, driving the entire analysis system to continuously develop higher precision and greater adaptability.

[0080] It should be noted that, in the present document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0081] While embodiments of the present application have been shown and described with reference to certain explanations, it is understood that those skilled in the art can make various changes, modifications, replacements and variations to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A real-time analysis system for airport video data, characterized in that, include: The video stream spatiotemporal feature modeling module is used to receive airport surveillance video stream data, perform multi-scale spatiotemporal feature analysis on continuous video frames, extract the trajectory segments of moving targets and the optical flow distribution of static background areas, identify the occurrence time sequence and spatial coordinates of key behavioral events, and generate a spatiotemporal behavioral feature sequence. The behavior trajectory map construction module, based on the spatiotemporal behavior feature sequence, performs hierarchical behavior modeling of the target motion trajectory and regional optical flow changes, establishes a topological connection path from macro region to micro target, collects the spatial correlation strength and temporal continuity attributes between path nodes, and outputs a multi-dimensional behavior trajectory map structure. The abnormal area association analysis module extracts the set of behavioral path nodes from different monitoring perspectives based on the multi-dimensional behavioral trajectory map structure, calculates the overlap of behavioral patterns of cross-regional nodes, filters the coordinates of frequently occurring endpoint areas, and forms an abnormal behavior area association set. The risk level semantic determination module obtains the predefined risk semantic labels for the corresponding regions based on the coordinates of the endpoint regions in the abnormal behavior region association set, sorts them by priority according to the frequency of regional risk association, matches the risk category of the endpoint region of the behavior trajectory, and generates a multi-level risk semantic label group. The situational structure visualization module maps the physical monitoring area nodes corresponding to each tag according to the multi-level risk semantic tag group, divides the behavioral trajectory data into area nodes according to risk level, constructs the spatiotemporal mapping relationship between area nodes and behavioral trajectories, and generates a heat map of the risk situation across the entire airport.

2. The airport video data real-time analysis system according to claim 1, characterized in that, The spatiotemporal behavioral feature sequence includes a set of moving target trajectory segments, a regional optical flow intensity distribution matrix, and a spatiotemporal coordinate index of behavioral events. The multi-dimensional behavioral trajectory map structure includes a hierarchical behavioral node mapping relationship, a spatiotemporal topological connection chain, and a set of node association strength parameters. The abnormal behavior region association set includes a set of cross-regional overlapping nodes, a frequency distribution of endpoint region coordinates, and the filtering results of associated regions. The multi-level risk semantic label group includes a risk label frequency sorting table, a endpoint region risk label mapping relationship, and risk level matching results. The airport's overall risk situation heat map includes regional node spatial identifiers, spatiotemporal grouping results of behavioral trajectories, and a risk situation mapping structure.

3. The airport video data real-time analysis system according to claim 1, characterized in that, The video stream spatiotemporal feature modeling module includes: The multi-scale feature parsing submodule acquires airport surveillance video stream data, uses a 3D convolutional network to decompose the spatiotemporal features of continuous video frames, separates the trajectory vector of moving targets from the background light flow field distribution matrix, records the starting frame number and spatial bounding box coordinates of each behavioral event on the time axis, calculates the ratio of the first trigger position of key behaviors in the spatiotemporal sequence to the number of continuous frames, and generates a spatiotemporal distribution index of behavioral events. The dynamic segment reconstruction submodule extracts video frame sequences before and after key behavior triggers based on the spatiotemporal distribution index of the behavior events, performs spatiotemporal alignment operations on motion trajectory vectors and optical flow matrices according to behavior types, and integrates multiple event segments under the same behavior category to form a set of spatiotemporal feature segments. The feature sequence generation submodule, based on the set of spatiotemporal feature segments, counts the trigger frequency of various behavioral events within a unit time window, rearranges the feature segment set according to the order of behavioral trigger time, splices multiple event segments of the same category into a continuous sequence according to the first trigger position, uses a dynamic time warping algorithm to calibrate the time offset of the behavioral feature sequence, and outputs a standardized spatiotemporal behavioral feature sequence.

4. The airport video data real-time analysis system according to claim 1, characterized in that, The behavior trajectory mapping construction module includes: The behavior hierarchy mapping submodule loads predefined airport area spatial topology weights based on the standardized spatiotemporal behavior feature sequence, sorts the motion trajectory nodes and optical flow area nodes by hierarchical priority, and establishes a node rearrangement index from the top area of ​​the terminal building to the last level area of ​​the boarding gate. The topology path generation submodule extracts the spatiotemporal correlation strength parameters between adjacent level nodes based on the node reordering index, records the bidirectional connection weight and temporal continuity index of each node, and uses a graph embedding algorithm to calculate the spatial density and temporal consistency coefficient of the node path to generate a spatiotemporal topology path with weight attributes. The graph structure extraction submodule, based on the spatiotemporal topological path with weighted attributes, collects the spatial coordinate connection relationship and temporal continuity parameters of all nodes in the path, and constructs a multi-dimensional behavioral trajectory graph structure containing a node spatial location mapping table and a temporal correlation matrix.

5. The airport video data real-time analysis system according to claim 1, characterized in that, The abnormal region correlation analysis module includes: The multi-view node acquisition submodule extracts the set of behavior path terminal nodes in the coverage area of ​​different cameras based on the multi-dimensional behavior trajectory map structure, and records the monitoring view number and path duration parameters corresponding to each terminal node. The cross-regional overlap analysis submodule, based on the set of terminal nodes, calls the spatial coordinate sequence of nodes from any two monitoring perspectives, calculates the overlap degree of node distribution in the spatial grid, extracts the frequently occurring endpoint coordinate points in the overlapping area, counts the frequency difference of such coordinate points in the multi-view path, and uses a spatiotemporal density clustering algorithm to filter the abnormal area coordinate set whose spatial clustering exceeds a preset threshold. The association set generation submodule, based on the abnormal area coordinate set, traces back the node number of the corresponding coordinate in the original behavior path, integrates the spatial mapping relationship between the monitoring view number and the area coordinate, and outputs an abnormal behavior area association set containing the area coordinate frequency distribution map.

6. The airport video data real-time analysis system according to claim 1, characterized in that, The risk level semantic determination module includes: The regional semantic mapping submodule, based on the coordinates of the endpoint region in the abnormal behavior region association set, queries the predefined risk semantic label library in the airport electronic fence database to establish an index mapping table between regional coordinates and risk labels; The risk frequency statistics submodule calculates the associated frequency value of each type of risk label in the abnormal area coordinate set according to the index mapping table, and generates a risk label priority sequence by sorting the frequency values ​​from high to low. The risk level matching submodule, based on the risk tag priority sequence, matches the risk tag category with the highest associated frequency for each abnormal area coordinate, integrates the tag matching results of all area coordinates, and generates a multi-level risk semantic tag group containing risk level weight values.

7. The airport video data real-time analysis system according to claim 1, characterized in that, The situation structure visualization module includes: The physical area mapping submodule, based on the multi-level risk semantic tag group, parses the airport physical area node number corresponding to each risk tag, and records the spatial coordinate range of the node in the electronic map and the number of existing risk trajectories. The trajectory space partitioning submodule, based on the node number of the airport physical area, groups and maps the spatiotemporal data of behavioral trajectories to the corresponding geographical areas according to the risk level weight, and establishes a two-way spatial index structure between trajectory data and geographical areas. The heatmap generation submodule, based on the bidirectional spatial index structure, calculates the risk trajectory density value within each geographical area, maps the density value to color gradient parameters, and overlays the airport electronic map to generate a visual heatmap layer containing the real-time risk intensity distribution.

8. The airport video data real-time analysis system according to claim 1, characterized in that, It also includes a real-time early warning feedback module, which performs the following operations based on the airport's overall risk situation heat map: Extract the set of coordinates of physical regions in the heat map whose risk intensity exceeds a preset threshold; Based on the multi-level risk semantic tag group associated with the physical area coordinate set, the priority of risk types is determined; Generate an early warning instruction data package containing the risk coordinates, risk type, and recommended handling measures; The warning instruction data packet is pushed to the responsible terminal equipment in the corresponding area in real time.

9. The airport video data real-time analysis system according to claim 8, characterized in that, The real-time early warning feedback module also includes: The early warning verification submodule receives the handling result data fed back by the responsible terminal; Compare the original risk intensity value of the area to be verified with the real-time risk intensity value after treatment; When the rate of decline in risk intensity fails to reach the preset target, the warning level will be automatically upgraded and a secondary response instruction will be triggered.

10. The airport video data real-time analysis system according to claim 3, characterized in that, It also includes an adaptive feature update module for performing: The rate of change of regional risk intensity in the heat map of the airport's overall risk situation is periodically collected. When the rate of change of risk intensity in a specific physical area continuously exceeds the dynamic threshold, a video stream feature re-extraction instruction for the corresponding area is triggered. Based on the re-extracted spatiotemporal behavioral feature sequence, the node association strength parameters in the multi-dimensional behavioral trajectory graph structure are reconstructed; Based on the reconstructed node association strength parameters, update the cross-regional behavior pattern overlap calculation model in the abnormal region association analysis module; The updated behavior pattern overlap calculation model is output to the risk level semantic determination module, and the generation logic of risk label priority sequence is optimized simultaneously.

Images