An intelligent security monitoring method and system, and a storage medium

Abstract

The application relates to the technical field of security monitoring, and discloses an intelligent security monitoring method and system and a storage medium, wherein the method is executed by a heterogeneous sensing joint verification processor in cooperation with a communication sensing fusion optical line terminal and other devices, a heterogeneous space-time mapping model of a fiber sensing link and a field of view of a network camera is constructed, environmental background noise features are extracted by using visual recognition results as true value labels to generate a background noise fingerprint library, in a real-time monitoring stage, the similarity of fiber vibration signals and noise fingerprints is calculated, and signal cleaning and intrusion judgment are carried out in combination with an adaptive threshold updating mechanism, in response to an alarm trigger signal, a quality of service scheduling is requested from a mobile communication core network to dynamically adjust the uplink bandwidth of the camera, and the camera positioning is controlled in linkage. Through the fusion verification and cross-layer resource scheduling of heterogeneous sensing data, the application solves the problems of high false alarm rate of fiber sensing and solidification of network resource allocation, and improves the accuracy and linkage efficiency of the system.

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology, specifically to an intelligent security monitoring method, system, and storage medium. Background Technology

[0002] Distributed fiber optic sensing technology is widely used in perimeter security, pipeline monitoring, and infrastructure protection due to its long-distance monitoring capabilities and resistance to electromagnetic interference. This technology typically utilizes Rayleigh backscattered signals in optical fibers to sense vibrations along the fiber, locating disturbance sources by analyzing changes in the phase or intensity of the optical signal. Meanwhile, video surveillance, as a direct verification method, is often used in conjunction with fiber optic sensing systems to provide visual evidence from the scene.

[0003] However, in practical applications, security systems relying solely on distributed fiber optic sensing technology face the problem of poor environmental adaptability. Because optical fibers are highly sensitive to vibration, weather factors such as wind, rain, and lightning, or even normal vehicle traffic near the perimeter, can generate significant vibration signals. Existing signal processing methods often rely on preset fixed thresholds or single-mode pattern recognition algorithms, making it difficult to effectively distinguish between such environmental background noise and genuine intrusion behavior, resulting in a high false alarm rate in complex environments.

[0004] When attempting to incorporate video surveillance for auxiliary verification, existing technologies generally suffer from difficulties in aligning heterogeneous data spatially. Fiber optic sensing provides one-dimensional distance information along the length of the fiber optic cable, while cameras provide two-dimensional images or three-dimensional spatial fields of view. There is a lack of a precise and automated spatiotemporal mapping mechanism between the two. This results in the system often being unable to quickly and accurately drive the camera to align with the specific vibration point for verification when the fiber optic cable detects vibration, or being unable to reverse-engineer visual information to mark noise samples in the fiber optic signal, thus limiting the deep fusion and collaborative analysis capabilities of heterogeneous sensing data.

[0005] Furthermore, in terms of data transmission and network resource allocation, existing security monitoring architectures typically employ static bandwidth allocation strategies. Network cameras, regardless of whether they are in monitoring or alarm mode, often occupy a fixed amount of uplink bandwidth. If configured in low-bandwidth mode, this may result in insufficient clarity of critical video evidence or transmission delays during intrusion events; conversely, if configured in high-bandwidth mode, it leads to a waste of communication resources in the absence of long-term anomalies. Current technology lacks a cross-layer scheduling mechanism that can dynamically request differentiated Quality of Service (QoS) guarantees or network slicing resources from the mobile communication core network based on the intrusion judgment results of the front-end sensing system. Summary of the Invention

[0006] To address the technical problems of existing distributed fiber optic sensing systems, such as high false alarm rates due to environmental noise interference, delayed linkage due to lack of accurate spatial mapping of heterogeneous devices, lack of cross-verification for single modes, and difficulty in ensuring communication quality during emergencies due to fixed network resource allocation, this invention provides an intelligent security monitoring method, system, and storage medium.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent security monitoring method, which is executed by a heterogeneous sensing joint verification processor in collaboration with a communication sensing fusion optical line terminal, a network camera, and a mobile communication core network, and includes the following steps: A spatiotemporal mapping model for heterogeneous sensing networks is constructed. By using the optical line terminal of the communication-sensing fusion optical fiber, the discrete sensing distance along the optical fiber is calculated using the principle of optical time-domain reflection. Combined with the reference vibration excitation applied within the field of view of the network camera, the location with the maximum response amplitude is identified. This establishes a mapping index between the physical location of the optical fiber sensing link and the geographical coordinates of the network camera's field of view, generating a heterogeneous spatiotemporal mapping database and solving the spatial alignment problem of heterogeneous data.

[0008] Perform vision-guided background noise fingerprint self-learning. Using the computer vision recognition results of network cameras as ground truth labels, perform frequency domain transformation on the fiber optic vibration phase data within the effective time window to extract environmental background noise features and generate fingerprints that are stored in the database, thereby identifying common environmental interference sources.

[0009] Real-time heterogeneous signal cleaning and intrusion detection are performed. The similarity between the real-time acquired fiber optic vibration signal features and the reference fingerprint in the background noise fingerprint database is calculated, and an adaptive threshold update mechanism based on statistical characteristics is used. When the similarity meets the high similarity condition, it is judged as background noise and cleaned; when the similarity meets the low similarity condition, it is judged as an abnormal intrusion event, thereby reducing the false alarm rate in complex environments.

[0010] Perform cross-layer resource scheduling and linked alarms. In response to the alarm trigger signal, send a service quality scheduling request to the mobile communication core network to apply for activation of network slice resources that ensure ultra-reliable low-latency communication, and dynamically adjust the uplink bandwidth of the network camera; at the same time, calculate the camera attitude parameters based on coordinate transformation algorithm and pinhole imaging principle, send positioning command to capture the target, and generate an alarm data packet containing multi-dimensional evidence and a unique digital fingerprint.

[0011] A second aspect of the present invention provides an intelligent security monitoring system applied to the above-described method.

[0012] The system includes: a communication-sensing fusion optical line terminal for extracting fiber vibration phase data based on Rayleigh backscattering; an optical distribution network as a distributed vibration sensor; a network camera for acquiring video streams and supporting dynamic bandwidth adjustment; and a heterogeneous sensing joint verification processor.

[0013] The heterogeneous sensing joint verification processor is configured to establish a heterogeneous spatiotemporal mapping database between the fiber optic link and the camera's field of view; generate a background noise fingerprint using visual results as truth labels; and perform signal cleaning or generate alarm trigger signals based on the similarity between the real-time signal and the fingerprint. The system also includes a command transmission device for sending resource scheduling requests and network slicing control signaling to the mobile communication core network; and an alarm device for processing alarm data packets.

[0014] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon.

[0015] When the computer program is executed by the processor, it implements the intelligent security monitoring method described in the first aspect above, including but not limited to the steps of constructing a heterogeneous spatiotemporal mapping model, performing background noise fingerprint self-learning, performing real-time signal cleaning and intrusion judgment, and performing cross-layer resource scheduling and linkage alarm.

[0016] This invention provides an intelligent security monitoring method, system, and storage medium. It offers the following advantages: 1. This invention achieves precise spatial alignment between fiber optic sensing and visual monitoring by constructing a heterogeneous spatiotemporal mapping model and combining it with visually guided noise self-learning. This mechanism can utilize truth labels to eliminate known environmental background noise, effectively solving the problem of high false alarm rates caused by environmental interference in traditional distributed fiber optic sensing systems, and significantly improving the spatial positioning accuracy and confidence of alarms.

[0017] 2. This invention employs real-time signal cleaning and an adaptive threshold update mechanism based on statistical characteristics, enhancing the system's environmental adaptability and detection robustness. By calculating signal similarity and dynamically adjusting the alarm threshold, the system can automatically distinguish between noise and intrusion signals in complex and ever-changing environments, overcoming the shortcomings of fixed threshold methods that are prone to missed or false alarms.

[0018] 3. This invention performs cross-layer resource scheduling and device linkage control, realizing dynamic matching between communication network resources and security service requirements. When an intrusion occurs, it automatically requests a high-priority network slice and guides the camera to track it, ensuring highly reliable retention of critical audio-visual evidence in emergency situations while effectively saving uplink bandwidth resources during routine monitoring. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system architecture of one embodiment of the present invention; Figure 2 This is a functional module logic block diagram of a heterogeneous sensing joint verification processor according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0020] Among them, 100 is the communication sensing fusion optical line terminal; 200 is the optical distribution network; 300 is the optical network unit; 400 is the network camera; 500 is the heterogeneous sensing joint verification processor; 501 is the memory; 510 is the spatiotemporal mapping construction module; 520 is the visual guidance feature extraction module; 530 is the heterogeneous signal cleaning module; 540 is the collaborative linkage control module; 600 is the command transmission device; 700 is the mobile communication core network; and 800 is the alarm device. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See attached document Figure 1 This invention provides an intelligent security monitoring system, which is built on a passive optical network architecture based on fiber optic broadband operation services and combined with a new generation of mobile communication core network for resource scheduling.

[0023] like Figure 1 As shown, the present invention provides an intelligent security monitoring system, which includes: Communication-sensing fusion optical line terminal 100: used to transmit communication optical signals downlink on the same optical interface and receive Rayleigh backscattered optical signals from the optical distribution network. By performing coherent detection and phase demodulation on the scattered optical signals, vibration phase data of the optical fiber along the line can be extracted.

[0024] Optical distribution network 200 and optical network unit 300: The optical distribution network 200 connects the communication sensing fusion optical line terminal 100 and the optical network unit 300 on the user side. While providing communication transmission for the optical network unit 300, it also serves as a distributed vibration sensor network.

[0025] Network Camera 400: Deployed within the coverage area of ​​the optical distribution network 200, it is used to collect real-time video image streams or extract characteristic data such as optical flow, and supports dynamic adjustment of its uplink bandwidth through network slicing strategies.

[0026] Heterogeneous sensing joint verification processor 500 and memory 501: Processor 500 is used to synchronously receive vibration phase data from communication sensing fusion optical line terminal 100 and video or feature data from network camera 400, and to make joint decisions based on the "heterogeneous spatiotemporal mapping database" recorded in memory 501. This database stores the mapping relationship between the fiber optic sensing location and the geographic coordinates of the camera.

[0027] Command sending device 600 and mobile communication core network 700: According to the decision of processor 500, command sending device 600 sends network slicing control signaling to mobile communication core network 700 to adjust the uplink transmission priority and service quality of network camera 400.

[0028] Alarm device 800: Used to receive alarm data packets containing multi-dimensional evidence output by processor 500, and to perform local audible and visual alarms or remote information push.

[0029] See attached document Figure 2 To achieve collaborative sensing and resource scheduling based on heterogeneous data, the heterogeneous sensing joint verification processor 500 contains multiple logical sub-modules. These modules are instantiated by the processor executing computer-readable instructions stored in memory 501. Specifically, the logical functional modules include a spatiotemporal mapping construction module 510, a visually guided feature extraction module 520, a heterogeneous signal cleaning module 530, and a collaborative linkage control module 540.

[0030] The spatiotemporal mapping construction module 510 is configured to establish a mapping table between the physical location of the fiber optic sensing link and the geospatial coordinates of the network camera 400. The spatiotemporal mapping construction module 510 first acquires the optical temporal reflectance data output by the communication-sensing fusion optical line terminal 100. Based on the time-of-flight principle of light pulses in optical fibers, let the speed of light in vacuum be... The effective refractive index of the fiber core is For the received bidirectional time interval is Calculate the corresponding fiber optic sensing distance from the Rayleigh backscattered signal. : ; In order to extend the one-dimensional fiber optic sensing distance Aligning with 3D geospatial space, the spatiotemporal mapping construction module 510 performs a calibration procedure: for specific geographic landmarks (with known coordinates) within the 400-degree field of view of the network camera. Apply artificial vibration excitation to identify the location with the largest response amplitude on the fiber optic link. This establishes a point-to-point mapping index. Through multi-point interpolation, a complete heterogeneous spatiotemporal mapping database is generated, which records the discrete sensing distance set. Coordinates of the center of the 400-degree field of view of the network camera The correspondence.

[0031] The visual-guided feature extraction module 520 is configured to perform optical flow analysis on video frame sequences from network cameras 400 to identify non-invasive moving targets in the environment. The visual-guided feature extraction module 520 constructs optical flow field constraint equations based on a constant brightness model. Let the image pixels... At any moment The brightness is Its instantaneous velocity components in the horizontal and vertical directions are respectively and Then it satisfies the partial differential equation: The visual guidance feature extraction module 520 calculates the average optical flow amplitude within the field of view. When this amplitude exceeds a preset motion threshold (e.g., the optical flow intensity corresponding to the normal driving speed of a vehicle) and the duration is within a set range, it determines that a legitimate moving target exists within the field of view. The visual guidance feature extraction module 520 records the start timestamp of this event. With end timestamp These two timestamps are generated based on a unified time synchronization protocol across the entire network (such as PTP or NTP) and are output as a valid time window.

[0032] The heterogeneous signal cleaning module 530 is configured to perform background noise fingerprinting and real-time signal filtering. Upon receiving a valid time window, the heterogeneous signal cleaning module 530 synchronously extracts vibration phase data output from the communication-sensing fusion optical line terminal 100 according to the time index. This module performs frequency domain transformation (e.g., Short-Time Fourier Transform, STFT) on the extracted time-domain data to generate a spectrum vector reflecting vehicle traffic characteristics and stores it in the background noise fingerprint database. During the real-time monitoring phase, the heterogeneous signal cleaning module 530 calculates the feature vector of the currently acquired real-time vibration signal. Reference fingerprint vector at the corresponding position in the background noise fingerprint database Euclidean distance between : ;in, The dimension of the feature vector. This refers to the dimension index within the feature vector; This represents the feature vector of the real-time acquired fiber optic vibration signal. ) in the Values ​​in each dimension; In the background noise fingerprint database, the first... One reference fingerprint vector ( ) in the The values ​​are calculated in each dimension. If the calculated minimum distance is less than the preset similarity threshold, the current vibration signal is determined to be known environmental background noise, and the heterogeneous signal cleaning module 530 performs a suppression operation, that is, discards the current alarm frame or marks it as "invalid". If the minimum distance is greater than or equal to the preset similarity threshold, it is determined to be an abnormal intrusion event, and a trigger signal is generated.

[0033] The collaborative control module 540 is configured to generate resource scheduling instructions for the mobile communication core network 700 in response to trigger signals. The collaborative control module 540 dynamically calculates the uplink communication bandwidth required by the network camera 400 based on the current intrusion judgment result. Assume the basic monitoring bandwidth is... The bandwidth of high-priority slices is Judgment status A value of 1 represents an intrusion, and a value of 0 represents normal operation. The allocation logic then satisfies: ; The coordinated control module 540 encapsulates the calculated bandwidth requirements into a service call request, which includes a 5G Quality of Service Identifier (5QI) or a Slice Differentiation Identifier (S-NSSAI). Finally, the request is sent to the network capability opening unit of the mobile communication core network 700 via the instruction sending device 600 to request the activation of network slice resources that guarantee ultra-reliable low-latency communication (URLLC).

[0034] The modules mentioned above are logically independent but tightly coupled in terms of timing. The spatiotemporal mapping construction module 510 provides a spatial reference for heterogeneous data alignment, the visual-guided feature extraction module 520 provides a truth label for environmental noise for the system, the heterogeneous signal cleaning module 530 uses this label to achieve physical layer suppression of false alarms from fiber optic sensing, and the collaborative linkage control module 540 realizes on-demand allocation of communication resources based on accurate decision results.

[0035] See attached document Figure 3 This invention provides an intelligent security monitoring method, which is jointly executed by a heterogeneous sensing joint verification processor, a communication sensing fusion optical line terminal, a network camera, and a mobile communication core network.

[0036] Step S100: Construct a spatiotemporal mapping model for the heterogeneous sensing network. In this step, the system establishes a mapping index between the physical location of the fiber optic sensing link and the geographic coordinates of the network camera's field of view. The communication-sensing fusion optical line terminal transmits probe light pulses to the optical distribution network and receives Rayleigh backscattered signals. Based on the bidirectional flight time of the light pulses in the optical fiber, the discrete sensing distance along the fiber is calculated. (i.e., the principle of optical time-domain reflectometry). Subsequently, the system applies a reference vibration at a known coordinate position within the network camera's field of view, identifies the location on the fiber optic link with the largest response amplitude, and then uses this discrete sensing distance... 3D geographic coordinates of the center of the network camera's field of view Establish bijective relations, generate and store a heterogeneous spatiotemporal mapping database.

[0037] Step S200: Perform vision-guided background noise fingerprint self-learning. In this step, the system uses computer vision recognition results as ground truth labels to extract legitimate background noise features in the environment. The network camera acquires video streams and uses optical flow to detect moving targets (e.g., normally driving vehicles) that conform to a preset motion pattern, generating a valid time window containing start and end timestamps. Based on a unified network clock, the system synchronously extracts corresponding discrete sensing distances according to the timestamp index. The vibration phase data at the location is used, and frequency domain feature transformation is performed on it to generate a background noise fingerprint vector. Stored in the dynamic fingerprint database to complete the registration of legitimate interference sources.

[0038] Step S300: Perform real-time heterogeneous signal cleaning and intrusion detection. During the real-time monitoring phase, when vibration signals are detected on the fiber optic link, the system extracts its real-time vibration feature vector. And calculate the vector and the corresponding reference fingerprint vector in the dynamic fingerprint database. The system calculates the similarity between the vibrations (e.g., Euclidean distance). The system compares the calculated result with a preset threshold: if the similarity is high (i.e., the distance is less than the threshold), the current vibration is determined to be registered environmental background noise, and signal cleaning (alarm masking) is performed; if the similarity is low (i.e., the distance is greater than or equal to the threshold), it is determined to be an abnormal intrusion event, and an alarm trigger signal is generated.

[0039] Step S400: Perform cross-layer resource scheduling and linked alarm. Once an alarm trigger signal is generated, the system immediately triggers the cross-layer linkage mechanism. The instruction transmitting device 600 sends a quality of service scheduling request to the mobile communication core network (such as PCF or NEF network elements) to dynamically adjust the uplink bandwidth resources allocated to the network cameras: maintaining the basic monitoring bandwidth under normal monitoring conditions. In the event of an intrusion decision, immediately switch to high-priority slice bandwidth. This ensures low latency and high reliability of video data. Simultaneously, the system generates alarm data packets containing vibration characteristics, video clips, and geographical location, and pushes them to the monitoring center.

[0040] For the principles of optical temporal reflection, the basic algorithms for optical flow calculation, and the basic protocol architecture for mobile communication network slicing involved in the above steps, those skilled in the art can refer to relevant communication and computer vision standards for implementation. Their basic principles are well-known technologies in this field and will not be elaborated here.

[0041] The following detailed description, in conjunction with specific embodiments, details the process of converting fiber optic transmission delay into geospatial coordinates in step S100 and establishing a mapping relationship with the visual sensor.

[0042] The communication-sensing fusion optical line terminal 100 is configured to periodically transmit highly coherent probe light pulses to the optical distribution network 200. During transmission through the optical fiber waveguide, the light pulses generate Rayleigh backscattered light, which carries phase information distributed along the fiber. The heterogeneous sensing joint verification processor 500 uses the time difference between the light pulse transmission time and the echo reception time for positioning.

[0043] Let the speed of light in a vacuum be... The effective refractive index of the fiber core is .definition To detect the bidirectional time interval between the emission of a light pulse and the receipt of a specific scattered signal. Based on the principle of optical time-domain reflectometry, the fiber optic sensing distance corresponding to this scattered signal is determined. for: ; The system performs analog-to-digital conversion on the received continuous analog signal, with a sampling rate of Continuous fiber optic sensing distance It is discretized into a series of continuous sensing units. The discrete sensing distance corresponding to each sampling point Calculations such as: ,in, For sampling index, .

[0044] To map the aforementioned discrete sensing distances to a geographic coordinate system, the heterogeneous sensing joint verification processor 500 executes a segmented spatial calibration procedure. This procedure includes the following steps: Step S211: Establish physical anchor point index. The system acquires GIS map data of the coverage area of ​​the network camera 400 and selects facilities with fixed geographical attributes (such as communication manholes, streetlights, or specific road markings) within the area as physical anchor points. Record the... Precise 3D geographic coordinates of each physical anchor point .

[0045] Step S212: Apply artificial excitation and acquire signals. At the physical anchor point At a specific geographical location, a transient vibration excitation with a specific frequency or intensity is applied (e.g., striking the road surface with a standard impact hammer). The communication-sensing fusion optical line terminal 100 synchronously acquires the Rayleigh backscatter signal sequence during this time period.

[0046] Step S213: Determine the fiber optic response location mapping. The heterogeneous sensing joint verification processor 500 calculates the distance of each discrete sensor. The differential energy value of the signal at that location. The system sets the signal-to-noise ratio threshold. The sampling index with the largest energy value exceeding the background noise benchmark and the largest amplitude was selected. The sampling index That is, physical anchor point The corresponding location on the fiber optic link. Based on this, the system generates mapping key-value pairs in the database. .

[0047] Step S214: Construct a piecewise linear interpolation model. Considering the excess length and route bends during optical cable laying, the system uses a piecewise linearization method to approximate the actual route of the optical cable. For two adjacent physical anchor points... (Corresponding index) )and (Corresponding index) The sensing units between the two sections, assuming the optical cable is uniformly distributed within the segment, for any index ( ), and its corresponding estimated geographic coordinates The calculation is as follows: This formula utilizes the proportional relationship between fiber optic distances to eliminate errors caused by discrepancies between physical and optical lengths due to excess cable length. The heterogeneous sensing joint verification processor 500 traverses all anchor point intervals, generating a heterogeneous spatiotemporal mapping database covering the entire link. In real-time monitoring, when the system detects... When an anomaly is detected, the coordinates are obtained by querying the database, and a PTZ control command is sent to drive the network camera 400 to adjust its azimuth and pitch angles to focus on the target area.

[0048] For the physical mechanism of Rayleigh scattering and the specific measurement of the effective refractive index of optical fibers, those skilled in the art can refer to relevant optical communication standards and testing specifications, which are well-known technologies in the field and will not be elaborated here.

[0049] This embodiment details the process by which the heterogeneous sensing joint verification processor 500 spatially registers the two-dimensional pixel coordinate system of the network camera 400 with the one-dimensional distance coordinate system of the fiber optic sensor using a computer vision algorithm. The process includes the following steps: Step S221: Construct the camera imaging model and obtain calibration parameters. In this step, the system establishes a pinhole imaging geometric model of the network camera 400. The heterogeneous sensing joint verification processor 500 reads the internal parameter matrix from the camera control interface. (Including focal length) and principal point coordinates The processor reads the current horizontal rotation angle in real time, along with external parameters. For cameras with fixed positions, external parameters are obtained through preset calibration. For pan-tilt-zoom (PTZ) cameras, the processor reads the current horizontal rotation angle in real time. Vertical pitch angle And the scaling factor. The system is based on the initial installation attitude matrix. Combined with rotation matrix operations Update the current external rotation matrix to achieve parameter tracking under dynamic field of view.

[0050] Step S222: Calculate the homography transformation matrix and its mapping to the ground plane. The system simplifies the mapping relationship in three-dimensional space to the ground plane ( Perspective transformation to the image plane. The heterogeneous sensing joint verification processor 500 retrieves physical anchor points as control points from the aforementioned heterogeneous spatiotemporal mapping database. Let the physical anchor points be... The coordinates in the world coordinate system are The corresponding coordinates in the image pixel coordinate system are The system selects at least four pairs of non-collinear control points and uses the least squares method to solve for the homography transformation matrix. Image pixel coordinates with world ground coordinates Satisfy projection relationship: ; ( (This is the scale factor). Through this matrix operation, the perspective transformation from physical coordinates to the image plane is achieved.

[0051] Step S223: Perform visual reprojection of the fiber optic link. In this step, the system traverses the discrete sensing distance set. For each discrete sensing distance The system queries its corresponding estimated geographic coordinates. The heterogeneous sensing joint verification processor 500 utilizes a homography transformation matrix. The estimated geographic coordinates Project back onto the image plane and calculate the coordinates of the center pixel. : ; The system determines the calculated center pixel coordinates. Does it fall within the image resolution range (i.e.) ,in Image width, (Image height). If it falls within the range, mark the discrete sensing distance. Mark it as a "visually visible node"; otherwise, mark it as a "visually blind spot node".

[0052] Step S224: Generate a visual region of interest (ROI) mapping table. To determine the specific pixel range corresponding to each node in the image, the system calculates the visually visible discrete sensing distance for each node. Generate a visual region of interest (ROI). The heterogeneous sensing co-verification processor 500 sets the physical monitoring width of the fiber optic monitoring. (For example, road width). The system constructs auxiliary geographic coordinate points. Its relationship with estimated geographic coordinates The physical distance is The system will assist with geographic coordinate points. Similarly, substituting the projection formula from step S223, we obtain the auxiliary pixel coordinates on the image. The system calculates the Euclidean distance between the center pixel coordinates and the auxiliary pixel coordinates to determine the pixel radius of the ROI region. The system uses center pixel coordinates. Centered on, with Construct a rectangular ROI mask for the side length This method ensures that the size of the ROI region can adaptively scale with changes in depth of field, accurately covering the actual physical space. The system constructs a lookup table in memory 501, with the following table entry structure: index key →Attribute value {estimated geographic coordinates} Center pixel coordinates ROI mask , Camera ID}.

[0053] Step S225: Handling occlusion and confidence update. To eliminate interference from tree or building occlusion, the system introduces a visual confidence coefficient. The heterogeneous sensing joint verification processor 500 performs Gaussian mixture background modeling (GMM) on the ROI region. The system counts the proportion of foreground pixels within the ROI region. If the region maintains a high proportion of static texture differences for a long period during periods of no fiber vibration (i.e., the background model cannot be updated to a clean road surface), or if the optical flow vector in the region remains zero for a long period during optical flow detection, the system determines that the node is "visually restricted." For "visually restricted" nodes, the system assigns a visual confidence coefficient... Set it to a low value (e.g., 0.1), and reduce the weight of the visual data for that node in subsequent joint verification; for normal nodes, set... The value is high (e.g., 1.0). When the PTZ camera rotates, causing a change in the field of view, the system responds to the position parameters transmitted back by the PTZ and re-triggers steps S222 to S225, dynamically updating the mapping table and the visual confidence coefficient.

[0054] For the fundamental theories of Singular Value Decomposition (SVD) in matrix operations and pinhole imaging models in computer vision, those skilled in the art can refer to relevant standard textbooks on computer vision, as these are well-known technologies in the field and will not be elaborated upon here.

[0055] This embodiment illustrates the specific process by which the heterogeneous sensing joint verification processor 500 extracts dynamic visual features corresponding to the spatial location of the fiber optic sensor from the video stream.

[0056] This process is executed by the heterogeneous sensing joint verification processor 500 and includes the following steps: Step S311: Video frame preprocessing and color space conversion. The network camera 400 transmits a video stream to the heterogeneous sensing joint verification processor 500 at a fixed frame rate. The processor stores the continuous video frame sequence in a buffer. Define variables. For the discrete-time index of the video stream (i.e., corresponding to the first...) (The acquisition time of the frame image). Assume it is at a discrete time. Image at pixel coordinates The gray value at that location is To suppress noise interference from the image sensor, the system performs Gaussian smoothing filtering on the grayscale image. The system constructs a discrete Gaussian convolution kernel with a size of 5×5. Its element weights are based on a two-dimensional Gaussian function. Generate, where, The standard deviation of the Gaussian distribution; Pi (a mathematical constant); This is the base of the natural logarithm (a mathematical constant). The system performs convolution operations. Output discrete time The smoothed grayscale image.

[0057] Step S312: Optical flow field calculation based on the assumption of constant brightness. In this step, the system calculates the motion velocity vectors of image pixels. Based on the assumption of constant brightness, constraint equations are established: ,in, , These are the velocity components in the horizontal and vertical directions, respectively. , These are the discrete spatial gradient values ​​in the horizontal and vertical directions obtained by convolving the image using the Sobel or Scharr operator. The discrete-time gradient is calculated as the difference between the gray values ​​at corresponding positions in the image at two adjacent discrete time points, i.e. To address the aperture problem, the system employs the Lucas-Kanade method, solving the equation using the least squares method within the local neighborhood of the pixel to obtain the velocity vector v. During the solution process, the processor evaluates the texture richness of a region by determining the minimum eigenvalue of the structure tensor matrix, and only performs this evaluation when the matrix is ​​invertible (eigenvalue greater than a threshold). Calculate optical flow when the velocity is zero, otherwise set the velocity to zero.

[0058] Step S313: Spatial Feature Aggregation Based on ROI Mask. The system, based on the previously generated Region of Interest (ROI) mapping table, performs spatial feature aggregation for each discrete sensing distance. Perform targeted feature extraction. System call index key. Corresponding ROI mask The mask defines the coordinates of the center pixel. The effective pixel range is used as a benchmark. The Heterogeneous Sensing Joint Verification Processor 500 calculates the ROI mask. Statistical characteristics of the optical flow velocity vectors of all pixels within the coverage area. The system calculates the average optical flow energy within this area. : ;in, ROI mask The total number of valid pixels contained within; Indicates at discrete time Below, image pixels In the horizontal direction ( The optical flow velocity component along the axial direction; Indicates at discrete time Below, image pixels In the vertical direction ( The optical flow velocity component along the axial direction. Furthermore, the system uses a two-parameter arctangent function to calculate the average motion direction angle. To obtain complete 360° orientation information: ; Step S314: Construct a sequence of visual feature vectors. The system combines the average optical flow energy with the motion direction angle, and incorporates the visual confidence coefficient determined in the aforementioned calibration stage. Generate discrete time intervals Below is a description of discrete sensing distance. Visual feature vectors : The visual feature vector It carries a precise timestamp and is stored in a heterogeneous feature cache queue. A visual confidence coefficient is introduced. Its function is to detect when a node is shaded or lacks sufficient light (i.e., When the visual feature amplitude of a node is relatively small, the system can automatically attenuate the amplitude of the node's visual features to prevent unreliable visual data from interfering with subsequent heterogeneous joint decisions.

[0059] For the specific operational logic of image convolution and the numerical algorithm of eigenvalue decomposition, those skilled in the art can refer to standard textbooks on digital image processing, as these are well-known technologies in the field and will not be elaborated upon here.

[0060] This embodiment details the process by which the heterogeneous sensing joint verification processor 500 acquires acoustic vibration information along an optical fiber and processes it into a digital feature sequence aligned with the visual signal in the time dimension.

[0061] The process mainly includes four stages: photoelectric signal conversion, coherent demodulation, phase restoration, and temporal feature aggregation. The specific execution steps are as follows: Step S321: Coherent optical signal detection and digitization. A narrow-linewidth laser within the communication-sensing fusion optical line terminal 100 operates at a repetition frequency... A sequence of probe pulses is transmitted. The variable `n` is defined as the sequence index of the probe pulses. The Rayleigh backscattered light returns along the optical fiber and enters the coherent receiver, where it interferes with the local oscillator light in a 90° optical mixer. The optical signal output from the optical mixer is converted into two orthogonal analog electrical signals, i.e., in-phase components, by a balanced photodetector. and orthogonal components Analog-to-digital converters (ADCs) operate at high sampling rates. Sampling is performed on both analog signals. For each discrete sensing distance... In each probe pulse index The system then obtains a pair of digitized orthogonal sampled values. and .

[0062] Step S322: Digital IQ Demodulation and Phase Extraction. The heterogeneous sensing joint verification processor 500 performs calculations on the digitized quadrature components to reconstruct the phase information of the light wave. For the detection pulse index n, located at the discrete sensing distance... The scattered signal at that location, its enveloping phase The calculation is as follows: ;in, It is a two-parameter arctangent function, and its output range is... .

[0063] Step S323: Phase unwrapping and vibration restoration. Because the wrapped phase is confined to... For a given interval, the system needs to perform an unwrapping operation to obtain the continuously changing absolute phase. The processor calculates the original phase difference between adjacent pulses. .like Then it is corrected to ;like Then it is corrected to The corrected phase difference is the time-difference phase. This physical quantity reflects the discrete sensing distance. The dynamic strain rate at each discrete sensing distance. Therefore, the system targets this for each discrete sensing distance. A one-dimensional vibration waveform sequence with a change in the probe pulse index n was obtained.

[0064] Step S324: Temporal feature aggregation and heterogeneous synchronization. Due to the fiber optic probe pulse repetition frequency... Much higher than the video frame rate of network cameras To achieve joint verification of heterogeneous data, the system needs to align the high-frequency fiber optic vibration signal to the discrete moments of the video stream. The system defines the aggregation window length. For the first video stream Each discrete-time index corresponds to a starting optical pulse index. The processor captures a segment of the fiber optic vibration waveform within this window. The heterogeneous sensing co-verification processor 500 calculates the fiber optic vibration energy within this window. : ;in, The aggregation window length (i.e., the total number of fiber optic probe pulses contained within one video frame period). The summation loop variable inside the aggregation window (within the range of 0 to 1). ); It is the time-difference phase (characterizing the strain rate of the optical fiber at a specific moment and location). To correspond to the current discrete time The index of the initial optical pulse; This is the index of the probe pulse sequence for the current calculation point within the aggregation window.

[0065] Simultaneously, in order to identify the type of vibration source, the processor performs a Fast Fourier Transform (FFT) on the waveform segment to obtain a frequency domain magnitude sequence. The system searches for the maximum value in the frequency domain magnitude sequence and defines the frequency point corresponding to this maximum value as the vibration center frequency. : ; Ultimately, the system generates discrete time intervals. Below is a description of discrete sensing distance. fiber optic feature vectors This vector is then stored in the heterogeneous feature cache queue. At this point, the vector is compared with the aforementioned visual feature vector. It achieved a dual alignment in both time and space.

[0066] For the balanced detection circuit structure and specific unpacking algorithm implementation code in coherent optical communication, those skilled in the art can refer to the Distributed Fiber Optic Sensing Technology Manual, which is a well-known technology in this field and will not be elaborated here.

[0067] This embodiment details how the heterogeneous sensing joint verification processor 500 uses the detection results of the vision subsystem to adaptively model the environmental background noise in the fiber optic sensing signal, thereby constructing a noise statistical model that can be dynamically updated with environmental changes. This process enables automatic learning and suppression of environmental interference even in the absence of target intrusion.

[0068] This process is executed periodically by the heterogeneous sensing joint verification processor 500, and specifically includes the following steps: Step S331: Silent State Determination Based on Visual Features. To obtain clean environmental noise samples, the system first confirms, based on visual data, that there are no moving targets in the physical space. The heterogeneous sensing joint verification processor 500 reads discrete time intervals from the cache queue. Below is a description of discrete sensing distance. Visual feature vectors The system extracts the average optical flow energy. and visual confidence coefficient The system sets a visual silence threshold. and confidence lower bound If the following logical conditions are met: Then determine the discrete sensing distance. At the current moment, the system is in a "visually silent state." This means that within the current time window, the visual detection results have not exceeded the preset motion threshold, and the vibration signal collected by the fiber optic sensor is marked as an environmental background noise sample.

[0069] Step S332: Screening and extraction of background noise samples. Once it is determined that the system is in a "visually silent state," the processor retrieves the corresponding fiber feature vector from the fiber feature cache queue. The system extracts the vibration energy from the optical fiber. As the current background noise sample If the above-mentioned silence condition is not met (i.e., a moving object is detected visually, or the confidence level is lower than the lower limit due to occlusion), the system will mark the data at the current moment as a "non-stationary sample". For non-stationary samples, the system will not perform model update operations, but will directly keep the statistical parameters of the previous moment unchanged to prevent vibration signals generated by intrusion from contaminating the noisy fingerprint database.

[0070] Step S333: Recursive Statistical Update and Fingerprint Modeling. The system uses the Exponentially Weighted Moving Average (EWMA) algorithm to recursively update the noise statistical characteristics. The system provides a fingerprint model for each discrete sensing distance... Maintain a set of noise statistics parameters: noise mean and noise variance During system initialization or the first fingerprint record creation, the system directly assigns the first valid background noise sample to the noise mean, i.e. The noise variance is initialized to 0 or a preset empirical minimum value. For subsequent discrete times... Using the selected background noise samples Update statistical parameters: ; ; in, The learning rate coefficient (value range 0 < <1, which determines the model's speed of following environmental changes. A larger [value] The smaller the value, the more the model can adapt to sudden weather changes. The value enhances the model's ability to smooth out transient disturbances.

[0071] Step S334: Generation and storage of dynamic adaptive threshold. Based on the updated statistical parameters, the system calculates the current threshold for the discrete sensing distance. Dynamic adaptive alarm threshold This threshold is composed of a noise basis plus a statistical bias term based on the statistical distribution: ;in, The sensitivity factor (usually ranging from 3 to 5, corresponding to the 3x3 of a Gaussian distribution) Or a higher probability range). The Heterogeneous Sensing Joint Verification Processor 500 will include The data set is defined as a "dynamic noise fingerprint" and updated to the heterogeneous spatiotemporal mapping database in memory 501.

[0072] Through the above steps, the system establishes a discrete sensing distance... This is a dynamic fingerprint database indexed by [the system]. When making subsequent intrusion decisions, the system will compare the real-time fiber optic vibration energy with this dynamic adaptive alarm threshold. This allows the system to automatically raise the alarm threshold during severe weather conditions such as heavy rain or strong winds, thereby significantly reducing the false alarm rate caused by environmental factors while maintaining the ability to detect strong intrusion behaviors.

[0073] For the specific numerical implementation of the exponentially weighted moving average algorithm and the data structure operations of computer memory, those skilled in the art can refer to relevant standard textbooks on data structures and algorithms, as these are well-known technologies in the field and will not be elaborated upon here.

[0074] This embodiment details how the heterogeneous sensing joint verification processor 500, during its online operation, utilizes a pre-built heterogeneous spatiotemporal mapping database to correlate independently collected fiber optic sensing data and visual data in real time along the temporal and spatial dimensions, generating a joint feature vector for intrusion determination.

[0075] This process is executed by the heterogeneous sensing joint verification processor 500 and includes the following steps: Step S411: Primary Fiber Screening Based on Adaptive Threshold. To improve processing efficiency, the processor uses a fiber optic sensing subsystem as the primary trigger source. At each discrete time... The system traverses all discrete sensing distances fiber optic feature vectors The system reads the distance for this discrete sensing distance. Dynamic adaptive alarm threshold This threshold is calculated using the statistical parameters maintained in the aforementioned dynamic noise fingerprint database generation step. The processor determines the fiber optic vibration energy at the current moment. Does it exceed the threshold? ; If the determination result is negative, the system ignores the node data; if the determination result is positive, the discrete sensing distance is recorded. It is marked as a "potential intrusion node" and triggers the subsequent cross-modal mapping process.

[0076] Step S412: Heterogeneous spatial index mapping and data alignment. This applies to discrete sensing distances marked as "potential intrusion nodes". The processor accesses the heterogeneous spatiotemporal mapping database in memory 501 and retrieves its associated visual attributes. The processor then... The corresponding spatial mapping relationship is searched in the heterogeneous feature cache queue for features with the same discrete time step. Visual feature vectors Considering potential packet loss or delays during network transmission, if the index is not found in the cache queue... For the corresponding visual feature vector, the system executes a missing data imputation strategy: constructing a visual feature vector filled with all zeros, and then adding the visual confidence coefficients from this vector... The value is forcibly set to 0. This strategy ensures that even if part of the vision subsystem fails, the system can still generate a unified joint feature vector for degradation processing, preventing program interruption.

[0077] Step S413: Construction and Normalization of Joint Feature Vectors. To eliminate differences between different physical dimensions, the system standardizes the extracted features. The Heterogeneous Sensing Joint Verification Processor 500 constructs a multidimensional joint feature vector. The system calculates the normalized fiber energy ratio. and visual energy ratio : ;in, The preset visual reference energy benchmark is set to the average optical flow energy value generated by a standard adult moving at a normal walking speed (e.g., 1.5 m / s) at a typical depth of field in a camera. Subsequently, the system generates a joint feature vector through vector concatenation. : ; In this vector, the visual confidence coefficient Included as a basis for subsequent classification algorithms to evaluate visual feature weights: when When the data is low (e.g., at night, due to occlusion, or missing data), subsequent classifiers will reduce the visual component. The weight of the judgment.

[0078] Step S414: Real-time mapping output of geospatial information. In addition to constructing feature vectors for classification, the system synchronously outputs the physical spatial location of the target. The processor directly reads the mapping database... Estimated geographic coordinates of the binding If the system ultimately determines that an intrusion event has occurred, the estimated geographic coordinates... It will be directly output as an alarm location parameter to drive the electronic map positioning display and linkage equipment control.

[0079] For key-value pair retrieval operations in databases and normalization algorithms for vector processing, those skilled in the art can refer to relevant standard textbooks on computer software engineering and data mining, as these are well-known technologies in the field and will not be elaborated upon here.

[0080] This embodiment illustrates how the heterogeneous sensing joint verification processor 500 utilizes a dynamic noise fingerprint database, combined with visual perception results, to perform a secondary statistical test on the potential intrusion node data initially screened. This process is used to eliminate fluctuating signals whose amplitude exceeds the initial threshold but conforms to the statistical distribution of background noise, as well as isolated interference caused by non-intrusive factors.

[0081] This process is performed by the Heterogeneous Sensing Co-verification Processor 500 for each discrete sensing distance that passes the initial screening. The execution specifically includes the following steps: Step S421: Statistical Normalization Score of Vibration Signal. To assess the significance of the current vibration signal relative to the historical environmental background, the system employs a statistical normalization method for measurement. The processor retrieves the discrete sensing distance from the heterogeneous spatiotemporal mapping database. Current corresponding noise statistics parameter: noise mean and noise variance The system uses these two parameters to measure the current fiber optic vibration energy. Perform standardization transformation and calculate the standard score (Z-score). : ;in, For a preset very small positive number (e.g., 10) −6 (This is used to prevent division by zero errors.) Standard score The standard deviation of the current signal from the center of the ambient background noise is quantitatively characterized.

[0082] Step S422: Calculation of false alarm suppression coefficient based on visual confidence. For unexplained signal abrupt changes (i.e., signal jumps without obvious external force) in fiber optic sensing, the system utilizes the vision subsystem for logical verification. The processor reads the visual confidence coefficient from the joint feature vector. The ratio of normalized visual energy The system defines a visual silence indicator function. If the conditions are met. ,but ,otherwise .in, The visual energy lower limit threshold is set to a value between 0.05 and 0.1. This value corresponds to the normalized energy level when only the inherent thermal noise of the image sensor exists in the video frame without any pixel displacement. This logic indicates that when the visual system has high confidence and confirms that there are no moving objects in the area, the current physical environment is determined to be silent. Based on this, the system calculates the false alarm suppression coefficient. : ;in, This is used to suppress the intensity factor (e.g., a value of 2.0). When the visually confirmed environment is silent, Increase, thereby raising the statistical significance requirement for fiber optic signals; when vision is obstructed or motion is detected, Keep it at 1 to maintain the original detection sensitivity.

[0083] Step S423: Weighted Cleaning Decision and Data Filtering. The system combines statistical scores and suppression coefficients to construct a cleaning decision inequality. The processor determines the standard score. Does it meet the following conditions: ,in, The statistical significance threshold is set between 3.0 and 5.0. If the inequality does not hold, the system determines that the signal belongs to random fluctuations in the environmental background or non-intrusive interference. The processor then sets the discrete sensing distance... The fiber vibration energy is marked as invalid data and will not be input into subsequent classifiers. If the inequality holds, it indicates that the signal is statistically significantly anomalous and has passed the visual logic verification; the system retains this joint feature vector. And then transmit it to the next-level processing module. Through this step, when the background noise increases and causes variance to increase... When it rises, the calculated This will decrease accordingly, thereby reducing the probability of false alarms; when visual confirmation indicates no target, by increasing... Raise the judgment threshold to filter out occasional interference.

[0084] For the statistical test principles of the normal distribution and the specific applications of the Z-score, those skilled in the art can refer to relevant textbooks on probability theory and mathematical statistics, as these are well-known techniques in the field and will not be elaborated upon here.

[0085] The above general process Figure 3 In step S400, “Execute cross-layer resource scheduling and linkage alarm”, when physical device control is involved, the heterogeneous sensing joint verification processor 500 performs the following detailed steps to drive the PTZ camera to aim at the target.

[0086] The process mainly includes three stages: coordinate inversion, zoom ratio determination, and closed-loop servo control. It is executed by the heterogeneous sensing joint verification processor 500 in collaboration with the network communication interface (command sending device 600). The specific steps are as follows: Target azimuth calculation and coordinate transformation. When the system generates a confirmed intrusion alarm event, the processor obtains the estimated geographic coordinates of the target. To drive the camera rotation, the system converts the Cartesian coordinates to polar coordinates relative to the camera itself. The system reads the installation position coordinates of the currently linked camera. and installation azimuth angle .in, The vertical height of the camera relative to the ground; The initial attitude deviation of the device is used to characterize this deviation, specifically defined as the clockwise angle between the camera's zero-position direction (usually the device's true north) and geographic true north. The processor calculates the required horizontal rotation angle (Pan) for the camera. And vertical pitch angle (Tilt) First, calculate the projected distance between the target and the camera on the horizontal plane. : ; Next, the absolute azimuth angle is calculated, and installation error correction and radian-to-degree conversion are performed to obtain the horizontal rotation angle. : ; Calculate the vertical pitch angle (With the horizontal line as 0 degrees and downward as negative): ; This calculation process establishes a precise mapping relationship between geospatial coordinates and camera mechanical motion coordinates, ensuring that the location detected by the fiber optic cable can be covered by the field of view of the visual sensor.

[0087] Adaptive zoom ratio calculation. To ensure the target has sufficient resolution in the video frame for identification, the system calculates the optimal optical zoom ratio. The system presets the target feature height. (Set at 1.7 meters, corresponding to the average adult height) and desired screen ratio (Set to 1 / 3 to 1 / 2 of the vertical screen). The target surface height of the camera's image sensor is known. and the base focal length of the current lens The processor calculates the required physical focal length of the target based on the pinhole imaging principle. : ;in, This is the straight-line distance. This allows us to obtain the optical zoom magnification required for the control commands. : ; Through this step, the system can automatically adjust the focal length according to the target distance, ensuring that the pixel density of the target in the image meets the requirements of the algorithm analysis.

[0088] Encapsulation of control signaling and vision-based closed-loop servoing. The processor will calculate the resulting set of parameters. The command is encapsulated into an absolute positioning command according to a standard monitoring protocol (such as ONVIF) and sent to the PTZ camera via the network communication interface (command transmitting device 600) to drive the pan-tilt unit to rotate to the coarse positioning point. After the camera is in position, the system enters "visual servo mode". The processor extracts the centroid coordinates of the moving target in the video frame in real time. And calculate its distance from the center point of the image. pixel deviation vector The processor generates proportional-derivative (PD) speed control instructions based on this deviation vector: ,in , This is the servo gain coefficient. The system continuously sends speed control commands. , Adjust the camera's orientation until the deviation vector converges to zero or the alarm is cleared. This ensures that the camera can continuously lock onto the target and keep it centered in the frame, even if the target is in motion.

[0089] For the specific message format of the ONVIF protocol and the PID control algorithm of the stepper motor, those skilled in the art can refer to the relevant communication standards of security monitoring systems and textbooks on automatic control principles. These are well-known technologies in the field and will not be elaborated here.

[0090] In the overall process step S400, in order to realize "service quality scheduling request" and "dynamic bandwidth allocation", the heterogeneous sensing joint verification processor 500, together with the network switching unit, performs the following detailed steps.

[0091] This process is executed by the heterogeneous sensing joint verification processor 500 cooperative network switching unit, and specifically includes the following steps: Definition and initial mapping of logical network slices. To achieve differentiated quality of service guarantees, the processor divides the physical network channel into three logical slices with different forwarding priorities: Signaling control slice (Slice-0): Used to transmit alarm messages and control commands, set to the highest priority (e.g., using EF to accelerate forwarding) to ensure low latency; Critical Data Slice (Slice-1): Used to transmit high-definition video streams from cameras identified as "intrusion states". It is set as the second highest priority (e.g., using AF41 forwarding) to ensure high throughput. General Surveillance Slice (Slice-2): Used to transmit video streams in a silent background state. It is set to low priority (e.g., using AF11 forwarding) to allow packet loss during congestion. During system initialization, a mapping table is established between camera identifiers (IDs) and logical slices. By default, all cameras are mapped to the general surveillance slice.

[0092] Real-time assessment of network congestion. The processor performs assessments at preset statistical intervals, at each discrete moment. Read the port traffic counter of the network switching unit and calculate the network load rate corresponding to that index. : ;in, This represents the total number of connected cameras. For the first Road camera at discrete time Real-time bit rate, Reserve bandwidth for signaling and protocol overhead. This represents the upper limit of the total physical bandwidth of the network link. If... Exceeding the preset congestion warning threshold (For example, if set to 80% of the link capacity), the system triggers a bandwidth reallocation process.

[0093] Dynamic weight calculation based on intrusion events. When the processor confirms the first... When an intrusion occurs in the coverage area of ​​a video camera, the slice mapping of that camera is immediately modified, elevating it to a "critical data slice." This is to calculate the target allocated bandwidth for each video stream. The system introduces dynamic weight coefficients. For hacking target cameras ,set up (High weight); for background cameras ,set up (Low weight). The system uses Weighted Fair Queuing (WFQ) logic to calculate and allocate bandwidth: ;in, Sum the indices for each camera (values ​​range from 1 to 1). ); To maintain the minimum keep-alive bitrate required for the video streaming connection, This is the effective available bandwidth after deducting signaling reservations. This formula ensures that all cameras maintain basic connectivity, while the remaining bandwidth resources are primarily allocated to cameras that capture intrusion targets, enabling them to transmit images in high definition.

[0094] Feedback adjustment of video encoding parameters. Calculated... The data is converted into control commands and sent to the video encoders of each camera to adjust the data output rate. For the first... The camera processor generates target bitrate control instructions. This instruction directs the camera's rate control to adjust its encoding target value: ,in, This represents the maximum encoding capability supported by the camera hardware. For background cameras categorized as part of the "General Surveillance Slice," the calculated allocated bandwidth... Approximately the lowest keep-alive bitrate The system simultaneously issues a "downgrade transmission command." This command includes two operations: first, reducing the video frame rate (FrameRate), and second, increasing the keyframe interval (I-frameInterval). By reducing the transmission frequency of the I-frame, which has the largest data volume, the average bandwidth usage of the background video stream is effectively compressed, thereby freeing up transmission resources for key segments.

[0095] Slice priority execution and reset. The network switching unit marks the DSCP field in the packet header according to the updated slice mapping table. When physical congestion occurs in the network, the switch scheduler strictly follows the order of Slice-0 > Slice-1 > Slice-2 for packet forwarding. When the intrusion alarm is cleared, the processor will... Weight of road cameras Reset to Then, it is remapped back to the "general monitoring slice", and the system bandwidth allocation is restored to an even distribution state.

[0096] For specific network device configurations of DSCP marking and Weighted Fair Queuing (WFQ), those skilled in the art can refer to the IEEE 802.1Q standard and related network management textbooks, as these are well-known technologies in the field and will not be elaborated upon here.

[0097] After completing the above resource scheduling and physical linkage, as the final output stage of step S400, the processor performs the following detailed steps to complete the alarm push.

[0098] This process is executed by the heterogeneous sensing joint verification processor 500 collaborative alarm management module, and specifically includes the following steps: Spatiotemporal alignment and structured encapsulation of multi-source evidence data. When an intrusion event is detected, the processor first obtains the current discrete time. The corresponding network synchronized clock generates standard Coordinated Universal Time (UTC). To construct a complete chain of evidence, the system extracts the following key data elements from the cache queue: Intrusion location vector That is, the geographical coordinates calculated above; Vibration characteristic segments Extract the fiber optic vibration energy sequence for a preset duration (e.g., from 3 seconds before to 3 seconds after) before and after the alarm time; Visual evidence anchors That is, the storage path of the target snapshot image captured by the linked camera; Overall confidence level This value represents the posterior probability of belonging to the "real intrusion" category after the aforementioned retained joint feature vector is input into a preset classifier (such as a support vector machine (SVM) or a fully connected neural network), and its value ranges from [0,1]. The processor constructs the above elements into a standardized alarm event vector. : ,in, This is a unique hardware identifier associated with the camera. Through this encapsulation, the system correlates abstract vibration waveforms with intuitive on-site images at the data level, enabling multi-dimensional forensics.

[0099] Calculation of data integrity check codes. To prevent alarm records from being tampered with during transmission or storage and to ensure the traceability of evidence, the system calculates alarm event vectors. Perform digital digest computation. The processor uses a secure hash algorithm (such as SHA-256) to generate a unique digital fingerprint. : ,in, This represents the operation of converting a data structure into a binary byte stream. This indicates a binary concatenation operation. This is a pre-set encryption salt value for the system. This digital fingerprint... It is appended to the end of the data packet and sent along with the alarm information. The receiving end recalculates the hash value using the same algorithm and then... Compare and verify the integrity of the data.

[0100] Priority routing strategy based on urgency. The system determines routing based on overall confidence level. Alarm events are categorized to determine the push notification strategy. A priority determination function is defined. : ;in, The emergency alarm threshold (e.g., 0.9). The warning threshold is set (e.g., 0.6). If it is determined to be... (Serious intrusion) The system triggers "multi-channel concurrent push mode," simultaneously sending data to the monitoring center's large screen, security personnel's handheld terminals, and external alarm interfaces; if determined to be... (Suspected loitering) Only writes logs to the backend database and sends a notification message to the on-duty personnel; if determined to be... The system discards the event or only records it in the local debug log, without consuming network bandwidth.

[0101] Adaptation and distribution of heterogeneous protocols. For different receiving terminals, the alarm management module will distribute structured data... The data is converted to a specific communication protocol format for transmission: For the monitoring center management platform, the system uses the WebSocket protocol to push the encapsulated JSON object to the front end in real time. For the mobile app, the system uses the MQTT protocol to publish the serialized data to the topic "alarm / critical / {Region_ID}" ({Region_ID} is the defense zone number). For historical archives, the system uses an HTTP POST request to store the complete chain of evidence in the database.

[0102] For the specific message specifications of the SHA-256 hash algorithm, JSON serialization format, MQTT and WebSocket communication protocols, those skilled in the art can refer to the IETF RFC standard documents and related network programming manuals, which are well-known technologies in the field and will not be elaborated here.

[0103] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described intelligent security monitoring method.

[0104] The storage media specifically include: Non-volatile memory (such as solid-state drives or flash memory): used to store compiled machine code instructions and system configuration files. It has internal data persistence logic to periodically or before system shutdown write back and save the updated "heterogeneous spatiotemporal mapping database" and dynamic noise fingerprints to ensure the system can recover its environment learning state after a power outage and restart. Volatile memory (such as random access memory): serves as the processor's runtime workspace, used to temporarily store high-frequency intermediate data during method execution, including real-time calculated standard scores. Alarm event vector And currently active dynamic fingerprint data, for the processor to read and write at high speed.

[0105] At the hardware execution level, the heterogeneous sensing joint verification processor 500 executes the following logic periodically in the form of a main loop by running the instructions in the aforementioned memory: Initial screening of fiber optic signals. The processor acquires real-time fiber optic vibration data and calculates a standard score by combining it with the retrieved noise fingerprint. If the score does not exceed the preset static alarm threshold, return directly to the loop start point; if it does exceed the threshold, proceed to the visual verification process.

[0106] Visual linkage verification. The processor calculates control parameters based on mapping relationships and drives the camera to turn towards the target area via a standard protocol (such as ONVIF). The instruction logic includes a status query or delay waiting mechanism to ensure that the visual algorithm (such as background subtraction or neural network) is called to perform secondary target confirmation in the specified area of ​​the image only after the pan-tilt-zoom motion has stabilized.

[0107] Alarm encapsulation and distribution. If visual verification confirms an intrusion, the processor performs a data aggregation operation to generate a structured alarm event vector containing spatiotemporal characteristics and cryptographic signatures (such as hash fingerprints). Based on the event priority, the vector is distributed to the monitoring terminal or storage terminal via an adapted network protocol (such as WebSocket or MQTT). The state is then reset and the processor returns to the main loop.

Claims

1. An intelligent security monitoring method, characterized in that, The method is jointly executed by a heterogeneous sensing joint verification processor, a communication sensing fusion optical line terminal, a network camera, and a mobile communication core network, and includes the following steps: Construct a spatiotemporal mapping model for a heterogeneous sensing network: Establish a mapping index between the physical location of the fiber optic sensing link in the optical distribution network and the geographic coordinates of the network camera's field of view, and generate a heterogeneous spatiotemporal mapping database; Perform vision-guided background noise fingerprint self-learning: using the computer vision recognition results of the network camera as the truth label, extract the environmental background noise features on the fiber optic sensing link, and generate a background noise fingerprint to be stored in the fingerprint database. Real-time heterogeneous signal cleaning and intrusion detection: During the real-time monitoring phase, the similarity between the feature vector of the real-time acquired fiber optic vibration signal and the reference fingerprint vector in the fingerprint database is calculated; if the similarity meets the preset high similarity condition, it is determined to be environmental background noise and signal cleaning is performed; if the similarity meets the preset low similarity condition, it is determined to be an abnormal intrusion event and an alarm trigger signal is generated. Perform cross-layer resource scheduling and linkage alarm: In response to the alarm trigger signal, send a service quality scheduling request to the mobile communication core network to dynamically adjust the uplink bandwidth resources of the network camera, and generate an alarm data packet containing multi-dimensional evidence.

2. The intelligent security monitoring method according to claim 1, characterized in that, The steps for constructing the spatiotemporal mapping model of the heterogeneous sensing network specifically include: The control communication sensing unit transmits probe light pulses and receives Rayleigh backscattered signals, and calculates the discrete sensing distance along the optical fiber based on the optical time domain reflection principle; A reference vibration excitation is applied to a known geographic coordinate location within the field of view of the network camera to identify the discrete sensing distance corresponding to the location with the largest response amplitude on the fiber optic link. A bijective relationship is established between the identified discrete sensing distance and the known geographic coordinates of the network camera's field of view center, and the heterogeneous spatiotemporal mapping database is generated using a piecewise linear interpolation method.

3. The intelligent security monitoring method according to claim 1, characterized in that, The steps for performing vision-guided background noise fingerprint self-learning specifically include: Perform computer vision analysis based on a constant brightness model on the video stream captured by the network camera to generate an effective time window containing a start timestamp and an end timestamp; According to the effective time window, vibration phase data at the corresponding discrete sensing distance is synchronously captured; frequency domain transformation is performed on the captured vibration phase data to generate a background noise fingerprint vector and store it in the background noise fingerprint database.

4. The intelligent security monitoring method according to claim 1, characterized in that, The steps for real-time heterogeneous signal cleaning and intrusion detection specifically include: Calculate the Euclidean distance between the current real-time vibration signal feature vector and the corresponding reference fingerprint vector in the background noise fingerprint database; The Euclidean distance is compared with a preset similarity threshold; If the Euclidean distance is less than a preset similarity threshold, it is determined to be of high similarity and confirmed as registered environmental background noise; If the Euclidean distance is greater than or equal to a preset similarity threshold, it is determined to be of low similarity and confirmed as an abnormal intrusion event.

5. The intelligent security monitoring method according to claim 1, characterized in that, The steps for real-time heterogeneous signal cleaning and intrusion detection also include an adaptive threshold update process based on statistical characteristics: Using the exponentially weighted moving average algorithm, the noise mean and noise variance are recursively updated based on the selected background noise samples; Based on the updated noise mean and noise variance, and combined with the preset sensitivity factor, the dynamic adaptive alarm threshold is calculated. During real-time decision-making, the real-time collected fiber vibration energy is compared with the dynamic adaptive alarm threshold to perform primary fiber screening.

6. The intelligent security monitoring method according to claim 1, characterized in that, The steps of performing cross-layer resource scheduling and linkage alarm specifically include the following processes involving physical device control: Obtain the estimated geographic coordinates corresponding to the abnormal intrusion event, and read the installation location coordinates and installation azimuth of the linked camera; The required horizontal rotation angle and vertical pitch angle of the camera are calculated based on a coordinate transformation algorithm. Based on the principle of pinhole imaging, the required optical zoom ratio is calculated according to the target feature height and the desired screen ratio. The calculated horizontal rotation angle, vertical pitch angle, and optical zoom ratio are encapsulated into positioning commands and sent to the network camera.

7. The intelligent security monitoring method according to claim 1, characterized in that, The steps of performing cross-layer resource scheduling and linkage alarm specifically include the following network resource scheduling process: Based on the current intrusion judgment, the required uplink communication bandwidth for the network camera is calculated using weighted fair queue logic. The calculated bandwidth requirements are encapsulated into a service call request that includes a slice differentiation identifier; The service call request is sent to the network capability opening unit of the mobile communication core network to request the activation of network slice resources that ensure ultra-reliable low-latency communication; The network resource scheduling process allocates high-priority slice bandwidth in the intrusion judgment state and basic monitoring bandwidth in the normal monitoring state.

8. The intelligent security monitoring method according to claim 1, characterized in that, The steps of performing cross-layer resource scheduling and linkage alarms specifically include the following alarm data processing procedures: Extract the intrusion location vector, vibration feature fragments, visual evidence anchor points, and comprehensive confidence level at the time of the alarm to construct a standardized alarm event vector; A unique digital fingerprint is generated by calculating the alarm event vector using a secure hash algorithm; The unique digital fingerprint is appended to the alarm data packet for integrity verification and distributed according to the priority strategy determined by the comprehensive confidence level.

9. An intelligent security monitoring system, characterized in that, The intelligent security monitoring method described in any one of claims 1-8 includes: The communication sensing fusion optical line terminal is used to send communication optical signals and receive Rayleigh backscattered optical signals to extract the vibration phase data of the optical fiber. An optical distribution network, connected to the communication-sensing fusion optical line terminal, serves as a distributed vibration sensor network. Network cameras are used to capture video image streams and support dynamic adjustment of uplink bandwidth; The heterogeneous sensing joint verification processor is connected to the communication sensing fusion optical line terminal and the network camera. It is used to establish a mapping index between the physical location of the optical fiber sensing link in the optical distribution network and the geographic coordinates of the network camera's field of view to generate a heterogeneous spatiotemporal mapping database. Using the computer vision recognition results of network cameras as truth labels, the environmental background noise features on the fiber optic sensing link are extracted and a background noise fingerprint is generated; and the similarity between the feature vector of the real-time acquired fiber optic vibration signal and the background noise fingerprint is calculated. Based on the similarity judgment result, signal cleaning or alarm trigger signal is generated. The instruction sending device is used to send a quality of service scheduling request to the mobile communication core network in response to the alarm trigger signal to dynamically adjust the uplink bandwidth resources of the network camera, and to send network slicing control signaling. An alarm device is used to receive alarm data packets containing multi-dimensional evidence and to trigger an alarm.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements any one of the intelligent security monitoring methods as described in claims 1 to 8.

Images