Multi-zone intelligent linkage alarm method based on AIoT gateway and related equipment

Abstract

The present application relates to a kind of based on AIoT gateway's multi-prevention area intelligent linkage alarm method and related equipment, including the following steps, the present application proposes a kind of based on AIoT gateway's multi-prevention area intelligent linkage alarm method, by the Internet of Things sensor environmental data of multiple prevention areas and video image stream are fused, generate prevention area scene three-dimensional reconstruction data, and it is monitored and analyzed, extract environmental characteristic sequence and trajectory analysis sequence.When detecting that multiple prevention area threat level exceeds threshold value, in combination with gateway historical alarm data, cause and effect inference is generated, generates cross-prevention area correlation event chain.Subsequently, matching AIoT gateway's preset collaborative response strategy, and through 4G / 5G network linkage SP voice call module, realize multi-channel alarm push, solve the technical problems that lack depth fusion and collaborative analysis means between video image data and environmental sensor data, it is difficult to form the overall understanding and dynamic modeling of prevention area scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of linkage alarm, in particular to a multi-zone intelligent linkage alarm method based on an AIoT gateway and related equipment. BACKGROUND

[0002] With the rapid development of Internet of Things and artificial intelligence technology, intelligent security systems are gradually evolving from single perception to multi-source heterogeneous data fusion. AIoT (Artificial Intelligence of Things) gateway, as the core node connecting edge devices and the cloud, has powerful local computing and collaborative control capabilities, providing a technical foundation for realizing multi-zone, cross-device intelligent linkage alarm. However, most current security systems still focus on independent zones, lacking effective identification and response mechanisms for potential associated events between multiple zones, leading to high false alarm rates and response lags in complex scenarios.

[0003] Existing video monitoring and sensor alarm systems often operate independently, lacking deep fusion and collaborative analysis methods between video image data and environmental sensor data, making it difficult to form a comprehensive understanding and dynamic modeling of the zone scene. Although some systems attempt to introduce three-dimensional reconstruction and behavior trajectory analysis technology, there are defects such as information loss and inaccurate feature matching in the multi-dimensional data fusion process, affecting the accuracy and real-time performance of threat identification. In addition, traditional alarm systems mainly rely on single threshold judgment mechanisms, making it difficult to respond to complex security threats such as multi-zone collaborative intrusion or distributed attacks.

[0004] More critically, existing intelligent security solutions generally lack causal reasoning mechanisms based on historical alarm data when facing cross-zone linkage events, failing to effectively uncover the internal connections between events, thereby limiting the ability to predict and respond to complex intrusion behaviors. At the same time, the push method of alarm information is relatively single, failing to fully utilize multi-channel resources such as voice call modules and mobile communication networks to achieve efficient linkage response. Therefore, there is an urgent need for a new intelligent alarm method that integrates multi-dimensional perception data, supports causal inference and strategy linkage, to improve the overall intelligence level and operational response capability of security systems. SUMMARY

[0005] The main purpose of the present application is to provide a multi-zone intelligent linkage alarm method based on an AIoT gateway, solving the technical problem of lack of deep fusion and collaborative analysis methods between video image data and environmental sensor data, making it difficult to form a comprehensive understanding and dynamic modeling of the zone scene.

[0006] To achieve the above purpose, the present application provides a multi-zone intelligent linkage alarm method based on an AIoT gateway, comprising the following steps:

[0007] The environmental data collected by the Internet of Things sensors of the multiple defense areas connected by the AIoT gateway are superimposed in the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion to generate defense area scene three-dimensional reconstruction data;

[0008] The defense area scene three-dimensional reconstruction data are subjected to correlation monitoring and identification analysis to obtain corresponding environmental feature sequences and trajectory analysis sequences;

[0009] When the threat level of any of the environmental feature sequences and the trajectory analysis sequences exceeds a preset threshold, the environmental feature sequences and / or the trajectory analysis sequences are subjected to causal inference based on the historical alarm data of the gateway to obtain a cross-defense area correlation event chain;

[0010] Based on the cross-defense area correlation event chain, a multi-defense area cooperative response strategy preset by the AIoT gateway is matched, and the multi-defense area cooperative response strategy is triggered through the preset 4G / 5G communication network to link to the SP voice call module to obtain a multi-channel alarm pushing result.

[0011] Further, the environmental data collected by the Internet of Things sensors of the multiple defense areas connected by the AIoT gateway are superimposed in the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion to generate defense area scene three-dimensional reconstruction data, comprising:

[0012] The environmental data collected by the Internet of Things sensors of the multiple defense areas connected by the AIoT gateway are subjected to multi-modal feature extraction to obtain environmental feature vectors, and the high-definition image stream is subjected to semantic segmentation through a deep convolutional neural network to obtain a scene semantic segmentation map; wherein the environmental feature vectors include temperature distribution features, humidity change features, air pressure fluctuation features, and light intensity features;

[0013] The environmental feature vectors and the scene semantic segmentation map are subjected to spatial alignment through a preset point cloud registration algorithm to obtain a multi-modal fusion data set, and the tensor decomposition technology is applied to the multi-modal fusion data set to obtain low-dimensional representation features;

[0014] Based on the low-dimensional representation features, a defense area scene semantic relationship network is constructed, and the defense area scene semantic relationship network is subjected to three-dimensional reconstruction through voxelization technology to obtain defense area scene three-dimensional reconstruction data.

[0015] Further, the defense area scene three-dimensional reconstruction data are subjected to correlation monitoring and identification analysis to obtain corresponding environmental feature sequences and trajectory analysis sequences, comprising:

[0016] The defense area scene three-dimensional reconstruction data are divided into multiple functional sub-regions, and the functional sub-regions are subjected to discrete representation through three-dimensional gridding processing to obtain a scene grid model;

[0017] perform time series analysis on the environment data of each grid unit in the scene grid model to obtain a grid unit time-varying feature matrix, and perform local correlation calculation on the grid unit time-varying feature matrix to obtain an environment feature sequence; wherein the environment feature sequence includes an abnormal hotspot distribution, a humidity abnormal diffusion path, and a light abnormal change region;

[0018] apply a three-dimensional target detection algorithm to the scene grid model to obtain a dynamic target set, and time-series correlate target motion trajectory data in the dynamic target set, wherein the target motion trajectory data includes a trajectory point sequence, a velocity vector, and acceleration information;

[0019] perform trajectory sequence classification on the target motion trajectory data to obtain a trajectory analysis sequence, wherein the trajectory analysis sequence includes a regular activity pattern, an abnormal behavior pattern, and a cross-region movement pattern.

[0020] Further, the environment feature sequence and / or the trajectory analysis sequence are causally inferred based on the gateway historical alarm data to obtain a cross-defense zone correlation event chain, including:

[0021] perform time series decomposition on the gateway historical alarm data to obtain an alarm event time series matrix, and perform abnormal pattern mining on the environment feature sequence and / or the trajectory analysis sequence based on the alarm event time series matrix to obtain an abnormal event feature;

[0022] based on the abnormal event feature, correlate map the environment feature sequence and / or the trajectory analysis sequence to obtain a multi-dimensional correlation feature spectrum, and construct a causal link based on the multi-dimensional correlation feature spectrum to obtain an event causal network;

[0023] perform probability propagation calculation on the event causal network to obtain an event state transition probability matrix, and perform dynamic threshold decision based on the event state transition probability matrix to obtain a cross-defense zone threat propagation sequence;

[0024] based on the cross-defense zone threat propagation sequence, perform spatio-temporal causal inference on the correlation events corresponding to the environment feature sequence and / or the trajectory analysis sequence in the preset multi-defense zone to obtain a cross-defense zone correlation event chain.

[0025] Further, the probability propagation calculation on the event causal network to obtain an event state transition probability matrix includes:

[0026] perform directed acyclic graph structure analysis on the topological relationship of the event nodes in the event causal network to obtain a node connection weight distribution, and apply a Bayesian inference mechanism to the node connection weight distribution to generate an initial conditional probability representation set;

[0027] propagation paths in the event causal network based on the initial condition probability representation set, to obtain an event propagation path set, and perform Markov chain forward deduction on the event propagation path set, to obtain a path state evolution sequence;

[0028] model the path state evolution sequence, construct a discrete-time state transition graph, and perform neighborhood influence analysis on the state transition process of each event node based on the discrete-time state transition graph, to obtain a node state transition local response function;

[0029] perform global state synchronization merging on all the event nodes based on the node state transition local response function, to obtain a global state transition graph, and perform probability matrix normalization processing on the global state transition graph, to obtain an event state transition probability matrix, wherein the event state transition probability matrix includes node-to-node state transition probability, intra-zone state propagation probability, and cross-zone state coupling probability.

[0030] Further, the propagation paths in the event causal network based on the initial condition probability representation set, to obtain an event propagation path set, include:

[0031] perform hierarchical decomposition on the initial condition probability representation set, to obtain a multi-level condition probability matrix, and perform path expansion on the inter-node relationship of the event causal network based on the multi-level condition probability matrix, to obtain a path candidate tree structure;

[0032] based on the path candidate tree structure, recursively calculate the propagation characteristics between event nodes, to obtain a node propagation feature vector group, and perform feature fusion on the node propagation feature vector group, to obtain a path propagation feature matrix;

[0033] perform weighted combination on the path propagation feature matrix, to obtain a path weight distribution graph, and perform optimal filtering on the event propagation paths in the event causal network based on the path weight distribution graph, to obtain an event propagation path set.

[0034] Further, the multi-zone collaborative response strategy matched by the cross-zone associated event chain based on the AIoT gateway is preloaded, and the multi-zone collaborative response strategy is triggered through the preset 4G / 5G communication network, to obtain a multi-channel alarm pushing result, including:

[0035] perform strategy analysis on the cross-zone associated event chain, to obtain an event response feature vector, and perform hierarchical mapping on the event response feature vector, to obtain a multi-level response decision matrix;

[0036] Match and screen the multi-zone cooperative response strategy preset by the AIoT gateway based on the multi-level response decision matrix, and perform conflict detection and exclusion on the multi-zone cooperative response strategy to obtain an executable strategy sequence;

[0037] Parallel decomposition is performed on the executable strategy sequence to obtain a parallel task graph, and resource scheduling optimization is performed based on the parallel task graph to obtain an execution scheduling scheme;

[0038] The execution scheduling scheme is triggered to the SP voice call module through the preset 4G / 5G communication network to obtain a multi-channel alarm pushing result.

[0039] The application also provides a multi-zone intelligent linkage alarm device based on an AIoT gateway, which comprises:

[0040] A fusion module is configured to superimpose environment data collected by Internet of Things sensors of multiple zones connected to the AIoT gateway on high-definition image streams collected by a video monitoring device in real time to perform multi-dimensional heterogeneous data fusion and generate three-dimensional reconstruction data of a zone scene.

[0041] An analysis module is configured to perform correlation monitoring and identification analysis on the three-dimensional reconstruction data of the zone scene to obtain an environment feature sequence and a trajectory analysis sequence.

[0042] An inference module is configured to perform causal inference on the environment feature sequence and / or the trajectory analysis sequence based on historical alarm data of the gateway when the threat level of any of the environment feature sequence and the trajectory analysis sequence exceeds a preset threshold to obtain a cross-zone correlation event chain.

[0043] A triggering module is configured to match the multi-zone cooperative response strategy preset by the AIoT gateway based on the cross-zone correlation event chain, and trigger the multi-zone cooperative response strategy to the SP voice call module through the preset 4G / 5G communication network to obtain a multi-channel alarm pushing result.

[0044] The application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of the above embodiments when executing the computer program.

[0045] The application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0046] This invention provides a multi-zone intelligent linkage alarm method based on an AIoT gateway, comprising the following steps: superimposing environmental data collected by IoT sensors in multiple zones connected to the AIoT gateway onto a high-definition image stream acquired in real time by a video surveillance device for multi-dimensional heterogeneous data fusion to generate three-dimensional reconstruction data of the zone scene; performing correlation monitoring and identification analysis on the three-dimensional reconstruction data of the zone scene to obtain corresponding environmental feature sequences and trajectory analysis sequences; when any of the environmental feature sequences and trajectory analysis sequences shows that the threat level of multiple zones exceeds a preset threshold, performing causal inference on the environmental feature sequences and / or the trajectory analysis sequences based on historical alarm data of the gateway to obtain a cross-zone correlation event chain; based on the... The cross-zone associated event chain matches the multi-zone collaborative response strategy pre-set in the AIoT gateway, and triggers the multi-zone collaborative response strategy to the SP voice call module through a preset 4G / 5G communication network, resulting in multi-channel alarm push results. This solves the technical problem of lacking deep integration and collaborative analysis methods between video image data and environmental sensor data, making it difficult to form a comprehensive understanding and dynamic modeling of the zone scene. It realizes the technical effect of effectively capturing the behavior patterns of people or objects in multiple zones by performing correlation monitoring and identification analysis on 3D reconstruction data, extracting environmental feature sequences and trajectory analysis sequences, thereby improving the accuracy of identifying potential threat events and reducing false alarms and missed alarms. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the steps of a multi-zone intelligent linkage alarm method based on an AIoT gateway in one embodiment of the present invention;

[0048] Figure 2 This is a structural block diagram of a multi-zone intelligent linkage alarm device based on an AIoT gateway in one embodiment of the present invention;

[0049] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0050] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] like Figure 1 As shown, Figure 1 This invention provides a multi-zone intelligent linkage alarm method based on an AIoT gateway, comprising the following steps:

[0053] Step S1, the environmental data collected by the Internet of Things sensors of the multiple defense zones connected by the AIoT gateway are superimposed in the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion to generate the three-dimensional reconstruction data of the defense zone scene.

[0054] Specifically, in this step, the Internet of Things sensors in the multiple defense zones connected by the AIoT gateway collect environmental data such as temperature, humidity, infrared, vibration, etc. in real time, and synchronize and fuse these data with the high-definition image stream collected by the video monitoring device, so as to realize the integration and analysis of multi-dimensional heterogeneous data. Since different types of sensor data and video images differ in time stamp, spatial coordinates and data format, it is necessary to perform unified time alignment and spatial mapping on these data by the AIoT gateway, so that each frame of video image contains not only visual information, but also environmental parameters from each sensor, forming an enhanced image stream with context semantics. On this basis, the system further utilizes a three-dimensional modeling algorithm to dynamically reconstruct the defense zone scene to generate a three-dimensional scene model with spatial depth and environmental characteristics. For example, in a perimeter security system, when an abnormal heat source is detected by an infrared sensor in a certain defense zone and a moving target is captured by a camera at the same time, the AIoT gateway can accurately identify the position, motion trajectory and accompanying environmental changes of the target after fusing the two types of data, providing comprehensive data support for subsequent behavior recognition and threat assessment.

[0055] Step S2, the defense zone scene three-dimensional reconstruction data are subjected to correlation monitoring and identification analysis to obtain environmental feature sequences and trajectory analysis sequences.

[0056] Specifically, in the process of correlation monitoring and identification analysis of the three-dimensional reconstruction data of the defense area scene, the system continuously monitors the multi-source information such as spatial structure information, environmental parameter changes and target motion state contained in the three-dimensional reconstruction data, and uses artificial intelligence algorithm to identify and model the dynamic behavior, thereby extracting the corresponding environmental feature sequence and trajectory analysis sequence. Specifically, the AIoT gateway combines the time dimension to analyze the environmental variables (such as temperature fluctuations, infrared heat source changes, vibration intensity, etc.) in multiple defense areas based on the three-dimensional scene model generated in the previous step, forming an environmental feature sequence; at the same time, for the targets (such as personnel, vehicles, etc.) identified in the video image stream, the moving path of the target in the three-dimensional space is obtained through target detection and tracking algorithm, and then a trajectory analysis sequence is generated. These two sequences together constitute the digital expression of the physical state and behavior pattern in the defense area scene, providing key basis for subsequent threat assessment and event causal inference. For example, in a perimeter security system, when the environmental feature sequence of a certain area shows abnormal temperature rise accompanied by increased vibration signal, and the trajectory analysis sequence identifies that a target crosses multiple defense areas and moves in an abnormal path, the system can preliminarily judge that there is a potential intrusion behavior, and the related data is transmitted to the next stage for further analysis and response.

[0057] Step S3, when any of the environmental feature sequence and the trajectory analysis sequence shows that the threat level of multiple defense areas exceeds the preset threshold, the environmental feature sequence and / or the trajectory analysis sequence are subjected to causal inference based on the historical alarm data of the gateway, to obtain a cross-defense area correlation event chain.

[0058] Specifically, when any of the environmental feature sequence and the trajectory analysis sequence shows that the threat level of multiple defense zones exceeds a preset threshold, the system will start the causal inference mechanism, based on the historical alarm data stored in the AIoT gateway, to conduct a deep correlation analysis on the current environmental feature sequence and / or trajectory analysis sequence, thereby identifying the potential causal relationship between events in different defense zones, and finally generating a cross-defense-zone correlated event chain. This process relies on the intelligent inference model built into the AIoT gateway, which learns from the event occurrence time, spatial distribution, behavior patterns, and other information in historical alarm data to establish logical connections between various abnormal events. When multiple defense zones currently have abnormal data, the model can automatically match historical patterns and deduce possible event evolution paths. For example, in the aforementioned perimeter security application scenario, if multiple adjacent defense zones' environmental feature sequences detect abnormal vibrations and infrared signals within a certain time period, and the trajectory analysis sequence identifies a moving target continuously crossing multiple defense zones with a characteristic of evading monitoring, the system will infer that there is a causal relationship between these abnormal behaviors by combining the occurrence sequence and response results of similar events in historical records, and integrate them into a cross-defense-zone correlated event chain, providing decision-making basis for the triggering of subsequent coordinated response strategies.

[0059] Step S4, based on the cross-defense-zone correlated event chain, the AIoT gateway preconfigured multi-defense-zone coordinated response strategy is matched, and through the preset 4G / 5G communication network, the multi-defense-zone coordinated response strategy is triggered to the SP voice call module, and a multi-channel alarm pushing result is obtained.

[0060] Specifically, in the process of matching the AIoT gateway preconfigured multi-defense-zone coordinated response strategy based on the cross-defense-zone correlated event chain, the system first matches the inferred event chain with various security response strategies preconfigured in the AIoT gateway, identifies the most suitable linkage disposal scheme for the current threat situation, and through the preset 4G / 5G communication network, the strategy instruction is real-time issued to the SP voice call module, thereby realizing the automatic triggering of the multi-channel alarm pushing mechanism. This step fully utilizes the edge computing capability and high-speed communication capability of the AIoT gateway, ensuring that when complex intrusion behaviors are detected, the corresponding alarm process can be quickly started, not only including traditional information pushing methods such as SMS and APP notification, but also realizing automatic voice broadcast through the SP voice call module, enhancing the instant communication effect of alarm information. For example, in the aforementioned perimeter security application scenario, when the system identifies that a moving target continuously crosses multiple defense zones accompanied by environmental abnormal signals, the AIoT gateway will match the corresponding emergency response strategy according to the generated cross-defense-zone correlated event chain, automatically call the SP voice call module to dial the security personnel's phone and play the preset voice alarm content, and at the same time, send graphic and text alarm information to the monitoring center through the 4G / 5G network, forming a multi-channel coordinated alarm, and improving the overall security response efficiency and accuracy.

[0061] In specific embodiments, the environmental data collected by the Internet of Things sensors of the multiple defense zones connected by the AIoT gateway are superimposed on the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion, generating defense zone scene three-dimensional reconstruction data, including:

[0062] Multi-modal feature extraction is performed on the environmental data collected by the Internet of Things sensors of the multiple defense zones connected by the AIoT gateway to obtain an environmental feature vector, and semantic segmentation of the high-definition image stream is performed through a deep convolutional neural network to obtain a scene semantic segmentation map; wherein the environmental feature vector includes temperature distribution features, humidity change features, air pressure fluctuation features, and illumination intensity features;

[0063] The environmental feature vector and the scene semantic segmentation map are spatially aligned through a preset point cloud registration algorithm to obtain a multi-modal fusion data set, and a tensor decomposition technique is applied to the multi-modal fusion data set to obtain low-dimensional representation features;

[0064] A defense zone scene semantic relationship network is constructed based on the low-dimensional representation features, and the defense zone scene semantic relationship network is three-dimensionally reconstructed using voxelization technology to obtain defense zone scene three-dimensional reconstruction data.

[0065] Specifically, in this step, the specific implementation process of "superimposing the environmental data collected by the IoT sensors of multiple defense zones connected by the AIoT gateway on the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion to generate defense zone scene three-dimensional reconstruction data" includes the following key technical links, and is completed through a series of algorithms and models. First, the AIoT gateway will interface with the Internet of Things sensor devices in multiple defense zones. These devices are distributed in different physical areas and are used to collect environmental data such as temperature, humidity, air pressure, and illumination. The system performs multi-modal feature extraction on these raw data to obtain environmental feature vectors with uniform format and dimension. This vector not only contains numerical information at the current time, but also incorporates time series trends such as temperature fluctuations over the past five minutes and humidity rise rates, thereby forming a richer context description. At the same time, the video monitoring device continuously collects high-definition image streams of each defense zone. To enable effective fusion with environmental data, the system uses a deep convolutional neural network (such as ResNet or U-Net) to perform semantic segmentation on each image frame, identifying key object classes and their spatial boundaries in the image, such as human forms, vehicles, fences, and vegetation, and outputting corresponding scene semantic segmentation maps. This semantic segmentation map annotates each object and its functional area in the scene at pixel-level precision, providing a structured foundation for subsequent spatial alignment. Next, the system performs spatial alignment of the environmental feature vectors and scene semantic segmentation maps using a pre-set point cloud registration algorithm (such as the ICP iterative closest point algorithm), ensuring that data from different sources can be mapped and fused in a unified three-dimensional coordinate system. For example, in a certain defense zone, an infrared sensor detects a heat source signal at a certain location, while a camera captures a moving target at the same orientation. The system uses the point cloud registration algorithm to accurately match these two types of information to the same spatial location, generating a multi-modal fusion dataset containing visual semantics and environmental parameters. This dataset not only records the appearance features of the target, but also includes the temperature, humidity, air pressure, and illumination of its location, making the description of the entire scene more comprehensive and three-dimensional. To further improve data processing efficiency and reduce the impact of redundant information, the system applies tensor decomposition techniques to the above multi-modal fusion dataset, converting high-dimensional raw data into low-dimensional representation features. In this process, the system organizes data from multiple sensors and cameras into high-order tensor form and performs dimensionality reduction using methods such as CP decomposition or Tucker decomposition to extract the most representative latent feature combinations. For example, within a certain time period, the system may extract a 200-dimensional environmental feature vector and a 300-dimensional visual feature vector from the raw data. After tensor decomposition, it can be compressed into a low-dimensional feature representation that retains only the first 50 main components, significantly reducing computational complexity while preserving core information.Based on these low-dimensional characterization features, the system further constructs a defense zone scene semantic relationship network. This network abstracts various objects, environmental factors, and their associated relationships in the scene into the form of nodes and edges through graph structure modeling. For example, a certain node may represent "personnel", and the edges connected to it may represent the environmental conditions (such as temperature, illumination) in which the personnel are located, their movement trajectories (connections to other nodes in previous and subsequent frames), and spatial relationships with other objects (such as access control, fences). This semantic relationship network not only reflects the structural information of the static scene, but also dynamically captures the behavior logic in the event evolution process. Finally, the system uses voxelization technology to perform three-dimensional reconstruction on the defense zone scene semantic relationship network, i.e., divides the entire defense zone into a number of voxel units (similar to three-dimensional pixels), each voxel unit contains the object types, environmental attributes, and dynamic behavior information within it, and thus constructs a complete three-dimensional space model. For example, in the application scenario of a perimeter security system, the system can divide the entire park into 1 meter x 1 meter x 1 meter voxel grids, each grid unit records whether there are personnel, vehicles, or other abnormal activities, and combines the temperature, humidity, infrared intensity, and other information provided by the surrounding sensors to form a fine three-dimensional visualization presentation of the entire defense zone. Through the above process, the system not only realizes data fusion from multiple defense zones and various types of devices, but also constructs a three-dimensional scene model with semantic understanding capability, providing a solid data support for subsequent threat monitoring, behavior analysis, and linkage response. For example, when a certain area experiences a sudden temperature rise, an increased infrared signal, and a video recognition of an unknown moving target, the system can accurately locate the event occurrence position in the three-dimensional reconstruction data and judge whether it constitutes a potential threat based on its historical behavior trajectory, thereby providing a reliable basis for intelligent alarm decision-making.

[0066] In specific embodiments, the correlation monitoring and identification analysis of the defense zone scene three-dimensional reconstruction data correspondingly obtains an environmental feature sequence and a trajectory analysis sequence, including:

[0067] The defense zone scene three-dimensional reconstruction data is divided into a plurality of functional sub-regions, and the functional sub-regions are discretely represented through three-dimensional gridding processing to obtain a scene grid model;

[0068] The environmental data of each grid unit in the scene grid model is subjected to time series analysis to obtain a grid unit time-varying feature matrix, and the grid unit time-varying feature matrix is subjected to local correlation calculation to obtain an environmental feature sequence; wherein the environmental feature sequence includes abnormal hot spot distribution, humidity abnormal diffusion path, and illumination abnormal change region;

[0069] applying a three-dimensional target detection algorithm to the scene mesh model to obtain a dynamic target set, and time-series correlating target motion trajectory data in the dynamic target set, wherein the target motion trajectory data comprises a trajectory point sequence, a velocity vector, and acceleration information;

[0070] performing trajectory sequence classification on the target motion trajectory data to obtain a trajectory analysis sequence, wherein the trajectory analysis sequence comprises a regular activity pattern, an abnormal behavior pattern, and a cross-region movement pattern.

[0071] Specifically, in this step, the specific implementation process of "correlation monitoring and identification analysis of the security area scene three-dimensional reconstruction data, corresponding to obtain the environmental feature sequence and trajectory analysis sequence" includes multiple technical links, and through space segmentation, time series modeling, target detection and trajectory classification, etc. Means to complete the deep mining of multi-dimensional security data. First, the system divides the security area scene three-dimensional reconstruction data generated in the previous step into multiple functional sub-regions according to physical functions or monitoring needs, such as the outer periphery of the fence in the park, the entrance channel, the equipment room and other space units with different purposes. In order to facilitate subsequent calculation and analysis, the system further performs three-dimensional grid processing on these functional sub-regions, which are discretized into several grid units with a voxel size of 1m x 1m x 1m, thereby constructing a clear and hierarchical scene grid model. On this basis, the system performs time series analysis on the environmental sensor data (such as temperature, humidity, air pressure, illumination, etc.) in each grid unit, and extracts its change trend in the continuous time period. For example, in a certain grid unit of a certain security area, the temperature is recorded to rise from 25°C to 38°C in the past 30 minutes, accompanied by a sharp drop in light intensity from normal value to zero. The system arranges these data into the time-varying feature matrix of the grid unit. Subsequently, the system uses local correlation calculation method (such as sliding window correlation coefficient analysis) to process the matrix, and identifies whether there is a synchronous change or causal relationship between adjacent grids. For example, if it is found that the temperature rise in a certain area is highly negatively correlated with the humidity drop in the east side area, it can be judged that there may be some heat exchange behavior between the two grids. Through this process, the system finally generates an environmental feature sequence containing abnormal heat spot distribution, humidity abnormal diffusion path and light abnormal change area, which is used to describe the dynamic environmental evolution of the entire security area. At the same time, the system also applies a three-dimensional target detection algorithm (such as PointPillars or VoxelNet) to the scene grid model, and identifies and locates the moving targets in the video monitoring and point cloud fusion data. This algorithm can accurately detect personnel, vehicles or other objects from three-dimensional space, and assign a unique ID to each detected target. Taking perimeter security as an example, when a person crosses the park fence and enters the internal area, the system will continuously track the target in multiple consecutive frames and record its position coordinates, motion direction, velocity vector and acceleration information in each grid unit, thereby forming a complete dynamic target set and its corresponding trajectory point sequence. In order to further understand the behavior pattern of the target, the system performs time series correlation analysis on the trajectory point sequence of all dynamic targets, and establishes the complete motion path of the target in space. For example, a target appears in grid A at minute 1, enters grid B at minute 2, and reaches grid C at minute 3, and its velocity vector shows that it is moving forward along a straight line at a high speed, and the acceleration is gradually increasing, which indicates that the target may be performing a break-in behavior.Through the statistics and modeling of a large amount of such trajectory data, the system can identify whether the target follows a regular activity pattern (such as a patrol personnel walking along a fixed route), an abnormal behavior pattern (such as lingering in an unauthorized area for a long time), or a cross-region movement pattern (such as crossing multiple defense zones in a short time). Based on the above analysis results, the system finally generates a trajectory analysis sequence, which contains the spatio-temporal distribution of various behavior patterns. For example, in an hour, the system detects 120 moving targets, of which 105 belong to the regular activity pattern, 10 exhibit abnormal behavior patterns, and 5 are identified as cross-region movement patterns. Combined with the abnormal hotspot distribution information in the environmental feature sequence, the system can further judge whether these behaviors are associated with environmental changes. For example, if the target of an abnormal behavior pattern appears in an area with a sudden temperature rise and enhanced infrared signal, the system can preliminarily determine that it has potential threat and transmit the relevant information to the next stage for causal inference and linkage response. In summary, through the spatial segmentation, grid modeling of the three-dimensional reconstruction data of the defense zone scene, time series analysis of environmental data, and trajectory recognition and classification of dynamic targets, the system can comprehensively capture various physical events and behavior patterns occurring in the defense zone, and generate structured environmental feature sequences and trajectory analysis sequences, providing a solid data foundation and logical support for subsequent threat assessment and intelligent linkage.

[0072] In specific embodiments, the causal inference of the environmental feature sequence and / or the trajectory analysis sequence based on the gateway historical alarm data obtains a cross-defense zone associated event chain, including:

[0073] The gateway historical alarm data is time-decomposed to obtain an alarm event time sequence matrix, and the environmental feature sequence and / or the trajectory analysis sequence is subjected to abnormal pattern mining based on the alarm event time sequence matrix to obtain abnormal event features;

[0074] Based on the abnormal event features, the environmental feature sequence and / or the trajectory analysis sequence is subjected to associated mapping to obtain a multi-dimensional associated feature map, and a causal link is constructed based on the multi-dimensional associated feature map to obtain an event causal network;

[0075] The event causal network is subjected to probability propagation calculation to obtain an event state transition probability matrix, and a dynamic threshold decision is made based on the event state transition probability matrix to obtain a cross-defense zone threat propagation sequence;

[0076] Based on the cross-defense zone threat propagation sequence, the associated events corresponding to the environmental feature sequence and / or the trajectory analysis sequence in the preset multi-defense zone are subjected to spatio-temporal causal inference to obtain a cross-defense zone associated event chain.

[0077] Specifically, in this step, the implementation process of "causal inference on the environmental feature sequence and / or the trajectory analysis sequence based on the gateway historical alarm data to obtain the cross-zone related event chain" involves multiple key algorithms and reasoning mechanisms. The core is to combine time series modeling, multi-dimensional feature mapping, and probabilistic reasoning to mine the potential event causal relationship between different zones and build a logical structure of the cross-zone related event chain. First, the system starts from the historical alarm data stored locally by the AIoT gateway, performs time series decomposition processing, extracts the periodic, trend and random fluctuation components, and generates an alarm event time series matrix. This matrix takes time as the axis and records the types, occurrence time, duration and intensity changes of various alarm events in each zone in the past period (e.g. the past 30 days). For example, in the application of a certain park security system, the system records that the 15th zone has occurred 8 times of infrared intrusion alarm in the past month, of which 6 times are concentrated in the night 21:00-24:00 period, and each alarm is accompanied by a sudden drop in light intensity and an abnormal rise in humidity. On this basis, the system uses the above alarm event time series matrix to perform anomaly pattern mining on the current environmental feature sequence and / or trajectory analysis sequence, and identifies the behavior characteristics or environmental change patterns highly related to historical alarms. For example, if the current environmental feature sequence of a certain zone shows that the temperature rises by 12°C in a short time, the humidity drops by 30%, and the trajectory analysis sequence detects a target crossing two adjacent zones at high speed, these features will be marked as abnormal event characteristics. By comparative analysis, the system can find out whether the current behavior is highly similar to a certain type of real threat event in history, thereby providing a basis for subsequent causal reasoning. Next, the system performs correlation mapping operation on the current environmental feature sequence and / or trajectory analysis sequence based on the extracted abnormal event characteristics, i.e. correlating and matching different dimensional data features in a unified semantic space to form a multi-dimensional correlation feature map. This map not only contains the environmental parameter changes in each grid cell, but also integrates the motion trajectory information of dynamic targets and their spatio-temporal adjacency relationship. For example, at a certain time, the system finds that a certain grid cell in A zone has a sudden temperature rise and an increase in infrared signal, while a fast-moving target is detected in the corresponding area of B zone, and the difference in time is only 2 minutes. This spatio-temporal consistency will be recorded as a potential event correlation. Subsequently, the system further constructs an event causal network based on the above multi-dimensional correlation feature map. The network adopts the form of directed graph, where nodes represent various abnormal events or behavior patterns, and edges represent the causal dependence relationship between events. For example, if the system observes that the event "temperature rise near the fence" always precedes "person climbing over the fence" multiple times, it will establish a causal link from the former to the latter in the causal network and assign a corresponding confidence weight.Through learning and statistics on a large amount of historical data, the system can continuously optimize the structure and parameters of the causal network, making it more accurately reflect the evolution law of events in the actual scene. In order to further quantify the propagation path and influence degree between events, the system performs probability propagation calculation on the event causal network to construct an event state transition probability matrix. This matrix describes the probability of the occurrence of other related events within a certain time in the future after a certain event occurs. For example, the system statistics found that when the "wall infrared alarm" occurs, there is a 72% probability that "personnel intrusion video screen" will occur within the next 5 minutes, and in 58% of the cases, it will eventually evolve into "illegal opening of access control". Based on these statistical data, the system can dynamically adjust the state transition probability between each event, so that the entire causal network has stronger prediction ability. On this basis, the system introduces a dynamic threshold judgment mechanism, which automatically adjusts the threshold standard for judging whether an event constitutes a threat according to the current environmental feature sequence and the change rate, amplitude and distribution density of the trajectory analysis sequence. For example, in the daytime with sufficient light and frequent personnel activities, the system may set a relatively high threshold for trajectory anomaly judgment; while in the night unattended period, the system will lower the threshold to increase the sensitivity to abnormal behavior. In this way, the system can adaptively identify real threat events in different time periods and different defense zone conditions. Finally, based on the above cross-defense zone threat propagation sequence, the system performs spatio-temporal causal inference on the pre-set environmental feature sequence and / or trajectory analysis sequence in multiple defense zones to identify a chain of related events that exist between multiple defense zones in terms of time sequence and spatial continuity. For example, in the application scenario of perimeter security, the system detects that a person triggers an infrared sensor in A defense zone, quickly crosses B defense zone and enters the equipment room of C defense zone, combined with historical data analysis, the system judges that this is an organized intrusion behavior, and abstracts it as a complete cross-defense zone related event chain: A defense zone infrared anomaly → A defense zone target appears → B defense zone rapid movement → C defense zone unauthorized entry → equipment room access control attempt to open. This event chain not only reveals the sequence of event occurrence, but also clearly shows the causal relationship between the behaviors at each stage, providing a key basis for subsequent coordinated response strategy formulation. In summary, by combining gateway historical alarm data, multi-dimensional feature mapping, causal network modeling and probability reasoning, the system realizes the deep understanding and intelligent identification of cross-defense zone events in complex security scenarios, effectively improves the perception and response ability to multi-point linkage threats, and lays a solid technical foundation for building a high-precision and intelligent security system.

[0078] In specific embodiments, the probability propagation calculation on the event causal network to obtain an event state transition probability matrix comprises:

[0079] topological relations of event nodes in the event causal network are analyzed by a directed acyclic graph structure to obtain a node connection weight distribution, and a Bayesian inference mechanism is applied to the node connection weight distribution to generate an initial condition probability representation set;

[0080] Based on the initial condition probability representation set, the event propagation path in the event causal network is path-weighted to obtain an event propagation path set, and Markov chain forward deduction is performed on the event propagation path set to obtain a path state evolution sequence;

[0081] State space modeling is performed on the path state evolution sequence to construct a discrete time state transition graph, and based on the discrete time state transition graph, a neighborhood influence analysis is performed on the state transition process of each event node to obtain a node state transition local response function;

[0082] Based on the node state transition local response function, global state synchronization merging is performed on all the event nodes to obtain a global state transition graph, and a probability matrix normalization process is performed on the global state transition graph to obtain an event state transition probability matrix, wherein the event state transition probability matrix includes node-to-node state transition probability, intra-zone state propagation probability, and cross-zone state coupling probability.

[0083] Specifically, in this step, the implementation process of the "probability propagation calculation of the event causal network to obtain the event state transition probability matrix" involves multiple key technical links from graph structure analysis, Bayesian inference modeling to global state synchronization merging, and the core goal is to transform the cross-defense zone event evolution process in a complex security scenario into a quantifiable probability representation through mathematical modeling means, thereby providing decision basis for subsequent dynamic threshold judgment and intelligent linkage response. First, the system analyzes the directed acyclic graph (DAG) structure of the event nodes in the event causal network constructed in the previous step, identifies the before-after dependency relationship between events, and extracts the connection weight distribution between each event node and other nodes. For example, in the application scenario of a certain park security system, the event causal network may include the following nodes: "wall infrared alarm", "target appears in A defense zone", "quickly crosses B defense zone", "tries to open C defense zone access control", etc. There is a clear order and causal direction between these nodes. The system finds through statistical analysis of historical alarm data that the connection weight from the "wall infrared alarm" node to the "target appears in A defense zone" node is 0.85, indicating that this path has a high confidence; while the connection weight from "quickly crosses B defense zone" to "tries to open C defense zone access control" is 0.72, indicating that this path is not absolutely occurring, but has certain regularity. Based on the above node connection weight distribution, the system further applies the Bayesian inference mechanism to generate an initial conditional probability representation set. Specifically, the system regards each event node as a random variable, whose value is "occurrence" or "non-occurrence", and estimates the probability of each node occurring under different premise conditions according to historical statistical data. For example, the probability of "target appears in A defense zone" occurring under the premise that "wall infrared alarm" has occurred is 91%, while under the premise that "wall infrared alarm" has not occurred, it only accounts for 3%. Thus, the system can establish a conditional probability table (CPT) for each event node, forming a complete Bayesian network structure model, laying a foundation for subsequent probability propagation deduction. Subsequently, the system performs path weighting expansion operation on the event propagation paths in the event causal network based on the initial conditional probability representation set, and identifies all possible event propagation path sets. Each path is composed of several consecutive event nodes and is assigned a corresponding path weight to represent the likelihood of the path occurring in the actual scenario. For example, the system identifies a typical propagation path: "wall infrared alarm → target appears in A defense zone → quickly crosses B defense zone → tries to open C defense zone access control", with a path weight of 0.45, indicating that this path has occurred 12 times in the past month, accounting for 45% of the total path number. The system sorts all paths by weight and selects high-frequency and high-risk propagation paths as the focus of subsequent deduction. On this basis, the system performs Markov chain forward deduction on the event propagation path set to simulate the evolution process of events on the time axis and generate path state evolution sequences.In this process, the system assumes that the current event state only depends on the state of the previous moment, which conforms to the Markov property. For example, after the "wall infrared alarm" occurs, the system predicts that the probability of "target appearing in A defense zone" at the next moment is 86%, and if the event does occur, the probability of "quickly crossing B defense zone" is 78%. In this way, the system can generate a series of path state evolution sequences and record the frequency of each state transition. Next, the system models the path state evolution sequence in the state space and constructs a discrete-time state transition graph. This graph takes time as the horizontal axis and state as the vertical axis, recording the state changes of each event node at different time steps. For example, at t=0, the "wall infrared alarm" is in the active state; at t=1, "target appears in A defense zone" is triggered; at t=2, "quickly cross B defense zone" starts to execute; at t=3, "try to open C defense zone access control" behavior appears. The system maps these states and their transition relationships to a unified state space to form a visual state transition graph. To better understand the local influence mechanism between events, the system further analyzes the state transition process of each event node based on the state transition graph, identifies the directly related pre-events and subsequent events, and generates a node state transition local response function accordingly. For example, the system analyzes and finds that the triggering of "quickly cross B defense zone" node is not only directly affected by "target appears in A defense zone", but also indirectly driven by factors such as the increase of ambient temperature (correlation coefficient 0.79) and the decrease of light intensity (correlation coefficient 0.68). Therefore, the system constructs a local response function for this node, which contains multiple input variables, to quantify its sensitivity to external disturbances. On this basis, the system performs global state synchronization and merging of all event nodes based on the node state transition local response function, that is, the state change process of all nodes in the entire event causal network is unified into the same time framework to form a global state transition graph covering the entire defense zone. For example, within a certain time period, the system detects the "target appears" event in A defense zone, the "quickly moves" event in B defense zone, and the "attempt to invade" event in C defense zone almost simultaneously, and combined with historical data analysis, it judges that this is a coordinated criminal behavior, which is merged into a complete event chain in the global state transition graph. Finally, the system normalizes the probability matrix of the global state transition graph to generate the event state transition probability matrix. This matrix includes three key probability indicators: first, the node-to-node state transition probability, such as the transition probability from "wall infrared alarm" to "target appears in A defense zone" is 0.89; second, the internal state propagation probability of the defense zone, such as the probability of "A defense zone target movement" to "A defense zone abnormal behavior" is 0.76; third, the cross-defense state coupling probability, such as the coupling probability of "B defense zone quickly crossing" to "C defense zone access control attempting to open" is 0.64.Through comprehensive analysis of these probability indicators, the system can accurately assess the propagation trend of events in the time and space dimensions, and provide a scientific basis for subsequent multi-channel alarm pushing and strategy linkage. For example, in the application scenario of perimeter security, the system detects that a target triggers an infrared sensor in A defense zone and quickly enters B defense zone, and approaches the C defense zone computer room within a short time. Through analysis of the event state transition probability matrix, the system judges that the behavior belongs to a high-risk propagation path, and the overall transition probability of "A defense zone infrared alarm" → "B defense zone rapid movement" → "C defense zone illegal approach" is as high as 0.83. Therefore, the linkage response mechanism is immediately started, the alarm instruction is sent to the SP voice module, and the monitoring center is pushed through the 4G / 5G network, so that the security personnel can respond at the best time and effectively prevent the escalation of security incidents.

[0084] In specific embodiments, the path weighting expansion of the event propagation path in the event causal network based on the initial condition probability representation set obtains an event propagation path set, including:

[0085] The initial condition probability representation set is hierarchically decomposed to obtain a multi-level conditional probability matrix, and the inter-node relationship of the event causal network is path expanded based on the multi-level conditional probability matrix to obtain a path candidate tree structure;

[0086] Based on the path candidate tree structure, the propagation characteristics between event nodes are recursively calculated to obtain a node propagation feature vector group, and the node propagation feature vector group is feature fused to obtain a path propagation feature matrix;

[0087] The path propagation feature matrix is weighted combined to obtain a path weight distribution diagram, and the event propagation path in the event causal network is optimally filtered based on the path weight distribution diagram to obtain an event propagation path set.

[0088] Specifically, in this step, the implementation process of "path-weighted propagation of event propagation paths in the event causal network based on the initial condition probability representation set" is a reasoning process from abstract probability model to specific event evolution path identification. This process relies on the initial condition probability representation set constructed by the Bayesian inference mechanism, and gradually excavates the propagation path set with high confidence in the event causal network through hierarchical decomposition, path expansion and weight calculation, etc. to provide key support for subsequent state transition modeling and intelligent response strategy matching. First, the system performs hierarchical decomposition on the initial condition probability representation set generated in the previous step, i.e. the joint probability relationship between complex event nodes is decomposed according to its hierarchical structure in the event causal network, forming a multi-level conditional probability matrix. These matrices are divided into multiple levels according to the depth of the causal link where the event node is located, for example, the first layer may contain basic trigger events (such as "wall infrared alarm"), the second layer may contain intermediate behaviors triggered by the first layer events (such as "target appears in A defense zone"), and the third layer may be the final threat behavior (such as "trying to open the access control"). The conditional probability matrix of each level records the probability distribution of the current layer event under different preconditions. For example, in the application of a certain park security system, the system statistics found that the probability of "target appearing in A defense zone" under the premise of "wall infrared alarm" is 91%, while the probability without it is only 3%; and the probability of "quickly crossing B defense zone" under the premise of "target appearing in A defense zone" is 78%. Through hierarchical organization of these data, the system can more clearly understand the stage characteristics of event propagation. Next, the system performs path expansion operation on the relationship between nodes in the event causal network based on the above multi-level conditional probability matrix, i.e. according to the directional connection relationship between nodes, it generates all possible event propagation path candidate tree structures. For example, assuming that a certain event causal network contains four nodes: A (wall infrared alarm), B (target appears in A defense zone), C (quickly cross B defense zone), and D (try to open C defense zone door), and there is a causal link A→B→C→D between them, the system will generate a complete path "A→B→C→D", and also identify other possible paths, such as "A→B→D", "A→C→D", etc. Each path represents a possible event evolution mode and constitutes part of the event propagation path candidate tree. This candidate tree structure not only covers the main path, but also includes various branch paths, reflecting the diversity and uncertainty of event propagation in actual security scenarios. Subsequently, the system performs recursive calculation on the propagation characteristics between event nodes based on the path candidate tree structure, extracts the dynamic influence factor of each node on different propagation paths, and generates a node propagation feature vector group.Each vector in the vector set corresponds to an event node and contains multiple dimensions of propagation feature parameters, such as "propagation delay time", "state transition intensity", "spatial diffusion range", etc. For example, in a certain intrusion behavior, the average propagation time from the "wall infrared alarm" node to the "target appears in A defense zone" node is 2.3 seconds, and the average propagation time from the "quickly cross B defense zone" node to the "attempt to open C defense zone access control" node is 4.5 seconds. These time information will be included in the feature vector as the propagation delay dimension. In addition, the system also analyzes the co-occurrence frequency, correlation coefficient and other statistical indicators of each node with other nodes in historical events to measure their influence and dependence in the entire propagation process. On this basis, the system further processes the node propagation feature vector set to convert it into a unified path propagation feature matrix. The matrix takes the path as the row and the feature dimension as the column, recording the comprehensive performance of each propagation path in each feature dimension. For example, for the path "A→B→C→D", its corresponding path propagation feature matrix may include the following values: "total propagation time = 8.6 seconds", "average transition intensity = 0.81", "number of cross-regional = 3", "environment disturbance times = 2", etc. These values constitute the overall propagation feature description of the path. In order to further filter out the propagation paths with high credibility, the system performs weighted combination operation on the path propagation feature matrix, assigns corresponding weight values according to the importance of different feature dimensions, and calculates the comprehensive path weight score of each path. For example, the system sets the propagation time weight as 0.3, the transition intensity weight as 0.4, the cross-regional number weight as 0.2, and the environment disturbance times weight as 0.1, and performs weighted summation on each path to obtain the path weight distribution graph. In a certain time period, the system detects that the weight score of the path "A→B→C→D" is 0.87, which is significantly higher than the average score of 0.56 of other paths, so this path is marked as a high-priority propagation path. Finally, the system filters and optimizes the propagation paths in the event causal network based on the path weight distribution graph, retains the top 20% of the weight ranking paths, and forms the final event propagation path set. For example, in the application scenario of perimeter security, the system filters out three main propagation paths "A→B→C→D", "A→B→D" and "A→C→D" after the above process, among which the first path has the highest weight, indicating that it is the most common and most threatening in the actual scene. These three paths together form the event propagation path set, providing accurate data support for subsequent state transition modeling, strategy linkage and risk assessment. In summary, through a series of operations such as hierarchical decomposition, path expansion, propagation feature extraction and weighted optimization from the initial condition probability representation set, the system realizes the fine identification and quantitative evaluation of the propagation paths in the event causal network. This mechanism not only improves the depth of understanding of the event evolution process, but also lays a solid foundation for subsequent multi-channel alarm pushing and collaborative response strategy execution.

[0089] In specific embodiments, the multi-zone cooperative response strategy pre-set by the AIoT gateway is matched based on the cross-zone associated event chain, and the multi-zone cooperative response strategy is triggered to the SP voice call module through a preset 4G / 5G communication network, to obtain a multi-channel alarm pushing result, including:

[0090] The event response feature vector is obtained by performing strategy analysis on the cross-zone associated event chain, and the multi-level response decision matrix is obtained by performing hierarchical mapping on the event response feature vector;

[0091] The multi-zone cooperative response strategy pre-set by the AIoT gateway is matched and screened based on the multi-level response decision matrix, and the executable strategy sequence is obtained by performing conflict detection and exclusion on the multi-zone cooperative response strategy;

[0092] The parallel task graph is obtained by performing parallel decomposition on the executable strategy sequence, and the execution scheduling scheme is obtained by performing resource scheduling optimization based on the parallel task graph;

[0093] The execution scheduling scheme is triggered to the SP voice call module through a preset 4G / 5G communication network, to obtain a multi-channel alarm pushing result.

[0094] Specifically, in this step, the implementation process of "matching the multi-zone cooperative response strategy preset by the AIoT gateway based on the cross-zone associated event chain, and triggering the multi-zone cooperative response strategy to the SP voice call module through the preset 4G / 5G communication network to obtain a multi-channel alarm pushing result" is a closed-loop intelligent response process from event recognition to strategy execution and alarm distribution. Relying on the powerful local computing power and edge decision mechanism of the AIoT gateway, combined with the high-concurrency transmission characteristics of the communication network, the rapid response and efficient disposal of cross-zone threat behaviors in complex security scenarios are realized. First, the system performs a strategy analysis operation on the cross-zone associated event chain generated in the previous step, that is, it converts the key event nodes, time sequence, spatial distribution, and causal dependence relationship in the event chain into a structured event response feature vector. For example, in the application scenario of perimeter security, if it is detected that a target continuously crosses A, B, and C three zones, accompanied by behaviors such as infrared anomaly, light mutation, and door access attempt opening, the system will extract response features including "intrusion path length", "cross-region number", "environment disturbance intensity", "target moving speed", etc. in multiple dimensions to form an event response feature vector containing 12 key indicators. Then, the system performs hierarchical mapping processing on the event response feature vector, compares it with the preset security level standard, and generates a multi-level response decision matrix. The matrix divides the event into three levels of low risk (Level 1), medium risk (Level 2), and high risk (Level 3) according to the threat level, and configures corresponding response strategy templates for each level. For example, when the event response feature vector shows "cross-region number = 3", "speed > 5m / s", "infrared signal enhancement amplitude = 60%", the system judges that the event belongs to Level 3 high-risk behavior, and maps it to the multi-level response decision matrix corresponding to the high-risk response level for subsequent strategy matching. Based on the above multi-level response decision matrix, the system starts to match and screen the multi-zone cooperative response strategy library preset by the AIoT gateway. The strategy library stores hundreds of pre-defined response rules, each rule corresponding to different event types, occurrence locations, impact ranges, and priorities. For example, for Level 3 intrusion behavior, the system can match a strategy named "three-level response - key area blockade and voice alarm", which includes starting A zone perimeter lighting, closing B zone entrance gate, activating C zone electronic fence, calling SP voice call module to dial security duty phone and playing preset alarm voice, etc. However, in actual application, there may be conflicts or resource contention problems among multiple response strategies. Therefore, the system further performs conflict detection and exclusion operation on the matched multi-zone cooperative response strategy to ensure that all instructions can be executed in coordination.For example, if one strategy requires opening the A-zone gate to release a drone patrol, and another strategy requires closing the A-zone gate to prevent personnel from entering, the system will identify the conflict and choose to retain the latter according to the preset priority rules, thus avoiding on-site execution confusion. After this stage, the system finally generates a set of conflict-free, executable strategy sequences for the next step of scheduling. Next, the system parallelizes the executable strategy sequences, breaking them down into multiple independently executable task units and constructing a parallel task graph. For example, the above-mentioned three-level response strategy can be decomposed into the following four parallel tasks: 1) start the A-zone lighting device; 2) close the B-zone gate; 3) activate the C-zone electronic fence; 4) call the SP voice call module to initiate an alarm call. These tasks have both partial time sequence dependency (such as the electronic fence should be activated before the target enters) and certain parallel execution conditions, so the system uses DAG (Directed Acyclic Graph) to model them and analyzes the execution order and resource occupation among the tasks. On this basis, the system optimizes the resource scheduling of the parallel task graph, taking into account factors such as the local computing load of the AIoT gateway, the available state of the sensor device, and the network bandwidth, to develop the optimal execution scheduling scheme. For example, at the current time, the AIoT gateway is running a video stream analysis task, and the CPU utilization has reached 75%, so the system decides to delay the execution of some non-real-time tasks (such as log recording) to prioritize the execution efficiency of critical actions such as alarm voice call and electronic fence activation. In addition, the system also predicts the execution time of each task based on historical data to ensure that the overall response time is controlled within 3 seconds, meeting the operational requirements. Finally, the system triggers the execution scheduling scheme to the SP voice call module through the preset 4G / 5G communication network, and simultaneously notifies other related devices to execute their respective tasks. For example, after receiving the alarm instruction, the SP voice call module immediately dials the on-duty phone of the security center and plays the preset voice: "Attention, illegal intrusion has occurred in C-zone, please handle immediately", while sending a text message and APP push information to relevant personnel. At the same time, the A-zone lights are automatically turned on, the B-zone gate is locked, and the C-zone electronic fence is powered on, and the entire security system completes all response actions within less than 2.5 seconds, significantly improving the emergency response efficiency. In summary, by extracting response features from cross-zone associated event chains, performing strategy matching and conflict resolution, task decomposition and scheduling optimization, and finally linking the SP voice call module through the 4G / 5G network, the system realizes intelligent, multi-channel, and multi-device collaborative response to complex security events. This mechanism not only improves the accessibility and timeliness of alarm information, but also enhances the overall disposal capability of cross-zone threats, providing solid technical support for building an efficient, safe, and intelligent security system.

[0095] The AIoT gateway-based multi-zone intelligent linkage alarm method in the embodiments of the present application is described above, and the AIoT gateway-based multi-zone intelligent linkage alarm device in the embodiments of the present application is described below. Please refer to Figure 2 One embodiment of the AIoT gateway-based multi-zone intelligent linkage alarm device in the embodiments of the present application includes:

[0096] The fusion module 21 is configured to superimpose the environmental data collected by the Internet of Things sensors of the multiple zones connected by the AIoT gateway in the high-definition image stream collected by the video monitoring device in real time to perform multi-dimensional heterogeneous data fusion, and generate zone scene three-dimensional reconstruction data.

[0097] The analysis module 22 is configured to perform correlation monitoring and identification analysis on the zone scene three-dimensional reconstruction data, and correspondingly obtain an environmental feature sequence and a trajectory analysis sequence.

[0098] The inference module 23 is configured to, when any of the environmental feature sequence and the trajectory analysis sequence shows that the threat level of the multiple zones exceeds a preset threshold, perform causal inference on the environmental feature sequence and / or the trajectory analysis sequence based on historical alarm data of the gateway, and obtain a cross-zone correlation event chain.

[0099] The triggering module 24 is configured to match the multi-zone cooperative response strategy preconfigured by the AIoT gateway based on the cross-zone correlation event chain, and trigger the multi-zone cooperative response strategy to the SP voice call module through a preset 4G / 5G communication network to obtain a multi-channel alarm pushing result.

[0100] In this embodiment, the specific implementation of each unit in the above device embodiment is described above in the method embodiment, and will not be described here.

[0101] Refer to Figure 3 In the embodiments of the present application, a computer device is also provided, and the internal structure of the computer device can be as shown in Figure 3 The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store the corresponding data in the embodiments. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0102] Those skilled in the art can understand that Figure 3The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied.

[0103] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the method.

[0104] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the method can be included. Any reference to memory, storage, databases, or other media in the present application and embodiments used herein can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0105] It should be noted that in this paper, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, device, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, device, article or method. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, device, article or method including the element.

[0106] The above merely describes preferred embodiments of the present application, and is not intended to limit the patent scope of the present application, and any equivalent structure or equivalent process conversion, or direct or indirect application in other related technical fields, which are made by using the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. An AIoT gateway-based multi-zone intelligent linkage alarm method, characterized in that, Comprise the following steps: The AIoT gateway connects the environment data collected by the Internet of Things sensors of multiple defense zones, and superimposes the high-definition image stream collected by the video monitoring device in real time to perform multi-dimensional heterogeneous data fusion, and generates defense zone scene three-dimensional reconstruction data; Correlation monitoring and identification analysis are performed on the defense zone scene three-dimensional reconstruction data, and corresponding environment feature sequences and trajectory analysis sequences are obtained; When any of the environment feature sequences and trajectory analysis sequences shows that the threat level of multiple defense zones exceeds a preset threshold, causal inference is performed on the environment feature sequence and / or the trajectory analysis sequence based on gateway historical alarm data, and a cross-defense zone correlation event chain is obtained; Based on the cross-defense zone correlation event chain, the multi-defense zone cooperative response strategy preset by the AIoT gateway is matched, and the multi-defense zone cooperative response strategy is triggered through the preset 4G / 5G communication network. SP voice call module linkage, get multi-channel alarm push result; The gateway historical alarm data is time-decomposed to obtain an alarm event time sequence matrix, and abnormal pattern mining is performed on the environment feature sequence and / or the trajectory analysis sequence based on the alarm event time sequence matrix to obtain abnormal event features. Based on the abnormal event features, the environment feature sequence and / or the trajectory analysis sequence are associated and mapped to obtain a multi-dimensional correlation feature spectrum, and a causal link is constructed based on the multi-dimensional correlation feature spectrum to obtain an event causal network; The event causal network is subjected to probability propagation calculation to obtain an event state transition probability matrix, and dynamic threshold judgment is performed based on the event state transition probability matrix to obtain a cross-defense zone threat propagation sequence; Based on the cross-defense zone threat propagation sequence, the correlation events corresponding to the environment feature sequence and / or the trajectory analysis sequence in the preset multi-defense zone are spatio-temporal causal inferred to obtain a cross-defense zone correlation event chain; The event causal network is subjected to probability propagation calculation to obtain an event state transition probability matrix, and dynamic threshold judgment is performed based on the event state transition probability matrix to obtain a cross-defense zone threat propagation sequence; The topological relationship of the event nodes in the event causal network is analyzed by directed acyclic graph structure to obtain a node connection weight distribution, and a Bayesian inference mechanism is applied to the node connection weight distribution to generate an initial condition probability representation set; Based on the initial condition probability representation set, the event propagation path in the event causal network is path-weighted and expanded to obtain an event propagation path set, and Markov chain forward deduction is performed on the event propagation path set to obtain a path state evolution sequence; The path state evolution sequence is subjected to state space modeling to construct a discrete time state transition graph, and the state transition process of each event node is subjected to neighborhood influence analysis based on the discrete time state transition graph to obtain a node state transition local response function; ​ The global state transition graph is obtained by globally synchronizing and merging all the event nodes based on the node state transition local response function, and the event state transition probability matrix is obtained by performing probability matrix normalization processing on the global state transition graph, wherein the event state transition probability matrix includes node-to-node state transition probability, internal state propagation probability in the defense area, and cross-defense area state coupling probability. 2.The AIoT gateway-based multi-zone intelligent linkage alarm method according to claim 1, characterized in that, The environmental data collected by the Internet of Things sensors of the multiple defense areas connected by the AIoT gateway are superimposed on the high-definition image stream collected by the video monitoring device in real time for multi-dimensional heterogeneous data fusion to generate defense area scene three-dimensional reconstruction data, including: The environmental data collected by the Internet of Things sensors of the multiple defense areas connected by the AIoT gateway are subjected to multi-modal feature extraction to obtain an environmental feature vector, and the high-definition image stream is subjected to semantic segmentation by a deep convolutional neural network to obtain a scene semantic segmentation map; wherein the environmental feature vector includes temperature distribution features, humidity change features, air pressure fluctuation features, and illumination intensity features; The environmental feature vector and the scene semantic segmentation map are spatially aligned by a preset point cloud registration algorithm to obtain a multi-modal fusion data set, and a tensor decomposition technique is applied to the multi-modal fusion data set to obtain low-dimensional representation features; A defense area scene semantic relationship network is constructed based on the low-dimensional representation features, and the defense area scene semantic relationship network is three-dimensionally reconstructed by voxelization technology to obtain defense area scene three-dimensional reconstruction data. 3.The AIoT gateway-based multi-zone intelligent linkage alarm method according to claim 1, characterized in that, The defense area scene three-dimensional reconstruction data is subjected to correlation monitoring and identification analysis to obtain an environmental feature sequence and a trajectory analysis sequence, including: The defense area scene three-dimensional reconstruction data is divided into multiple functional sub-regions, and the functional sub-regions are discretely represented by three-dimensional gridding processing to obtain a scene grid model; Time series analysis is performed on the environmental data of each grid cell in the scene grid model to obtain a grid cell time-varying feature matrix, and local correlation calculation is performed on the grid cell time-varying feature matrix to obtain an environmental feature sequence; wherein the environmental feature sequence includes abnormal hot spot distribution, humidity abnormal diffusion path, and illumination abnormal change region; A three-dimensional target detection algorithm is applied to the scene grid model to obtain a dynamic target set, and target motion trajectory data in the dynamic target set are time-series correlated, wherein the target motion trajectory data include trajectory point sequence, velocity vector, and acceleration information; The target motion trajectory data are subjected to trajectory sequence classification to obtain a trajectory analysis sequence, wherein the trajectory analysis sequence includes a regular activity mode, an abnormal behavior mode, and a cross-region movement mode. 4.The AIoT gateway-based multi-zone intelligent linkage alarm method according to claim 1, characterized in that, The event propagation path set is obtained by path weighting expansion of the event propagation path in the event causal network based on the initial condition probability representation set, including: The initial condition probability representation set is hierarchically decomposed to obtain a multi-level conditional probability matrix, and the path expansion of the node relationship in the event causal network is performed based on the multi-level conditional probability matrix to obtain a path candidate tree structure; Based on the path candidate tree structure, the propagation characteristics between the event nodes are recursively calculated to obtain a node propagation feature vector group, and feature fusion is performed on the node propagation feature vector group to obtain a path propagation feature matrix; The path propagation feature matrix is combined by weighting to obtain a path weight distribution diagram, and based on the path weight distribution diagram, the event propagation paths in the event causal network are filtered to obtain an event propagation path set. 5.The AIoT gateway-based multi-zone intelligent linkage alarm method according to claim 1, characterized in that, The multi-zone collaborative response strategy pre-stored in the AIoT gateway is matched based on the cross-zone associated event chain, and the multi-zone collaborative response strategy is triggered to the SP voice call module through the preset 4G / 5G communication network, to obtain a multi-channel alarm pushing result, including: The multi-zone collaborative response strategy pre-stored in the AIoT gateway is matched based on the multi-level response decision matrix, and the multi-zone collaborative response strategy is subjected to conflict detection and exclusion to obtain an executable strategy sequence; The multi-zone collaborative response strategy pre-stored in the AIoT gateway is matched based on the multi-level response decision matrix, and the multi-zone collaborative response strategy is subjected to conflict detection and exclusion to obtain an executable strategy sequence; The multi-zone collaborative response strategy pre-stored in the AIoT gateway is matched based on the multi-level response decision matrix, and the multi-zone collaborative response strategy is subjected to conflict detection and exclusion to obtain an executable strategy sequence; The multi-zone collaborative response strategy pre-stored in the AIoT gateway is matched based on the multi-level response decision matrix, and the multi-zone collaborative response strategy is subjected to conflict detection and exclusion to obtain an executable strategy sequence.

6. An AIoT gateway-based multi-zone intelligent linkage alarm device, characterized in that, It includes: The fusion module is used for superimposing the environment data collected by the Internet of Things sensors of the multiple zones connected by the AIoT gateway on the high-definition image stream collected by the video monitoring device in real time to perform multi-dimensional heterogeneous data fusion and generate zone scene three-dimensional reconstruction data; The analysis module is used for performing correlation monitoring and identification analysis on the zone scene three-dimensional reconstruction data to obtain environment feature sequences and trajectory analysis sequences; The inference module is used for performing causal inference on the environment feature sequences and / or the trajectory analysis sequences based on the gateway historical alarm data when the threat level of any of the environment feature sequences and the trajectory analysis sequences exceeds a preset threshold to obtain a cross-zone associated event chain; The trigger module is used for matching the multi-zone collaborative response strategy pre-stored in the AIoT gateway based on the cross-zone associated event chain, and triggering the multi-zone collaborative response strategy to the SP voice call module through the preset 4G / 5G communication network to obtain a multi-channel alarm pushing result; The inference module is used for performing causal inference on the environment feature sequences and / or the trajectory analysis sequences based on the gateway historical alarm data to obtain a cross-zone associated event chain, including: The gateway historical alarm data is subjected to time sequence decomposition to obtain an alarm event time sequence matrix, and based on the alarm event time sequence matrix, abnormal pattern mining is performed on the environment feature sequences and / or the trajectory analysis sequences to obtain abnormal event features; Based on the abnormal event characteristics, the environment feature sequence and / or the trajectory analysis sequence are associatedly mapped to obtain a multi-dimensional association feature graph, and a cause-effect link is constructed based on the multi-dimensional association feature graph to obtain an event cause-effect network; An event state transition probability matrix is obtained by performing probability propagation calculation on the event cause-effect network, and a dynamic threshold decision is made based on the event state transition probability matrix to obtain a cross-defense zone threat propagation sequence; Based on the cross-defense zone threat propagation sequence, a spatio-temporal cause-effect inference is performed on the associated events corresponding to the environment feature sequence and / or the trajectory analysis sequence in the preset multi-defense zone to obtain a cross-defense zone associated event chain. The probability propagation calculation on the event cause-effect network to obtain the event state transition probability matrix includes: The topological relationship of the event nodes in the event cause-effect network is analyzed by a directed acyclic graph structure to obtain a node connection weight distribution, and a Bayesian inference mechanism is applied to the node connection weight distribution to generate an initial condition probability representation set; Based on the initial condition probability representation set, the event propagation path in the event cause-effect network is path-weighted expanded to obtain an event propagation path set, and a Markov chain forward deduction is performed on the event propagation path set to obtain a path state evolution sequence; The path state evolution sequence is state space modeled to construct a discrete time state transition graph, and based on the discrete time state transition graph, a neighborhood influence analysis is performed on the state transition process of each event node to obtain a node state transition local response function; Based on the node state transition local response function, a global state synchronization merging is performed on all the event nodes to obtain a global state transition graph, and a probability matrix normalization processing is performed on the global state transition graph to obtain an event state transition probability matrix, wherein the event state transition probability matrix includes node-to-node state transition probability, defense zone internal state propagation probability, and cross-defense zone state coupling probability. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6. The processor executes the computer program to realize the steps of the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method of any one of claims 1 to 5.

Images