A door access video monitoring linkage management method and system

Abstract

The application relates to the field of security monitoring, and discloses a door access video monitoring linkage management method and system. The method comprises the following steps: acquiring a door access triggering event and associated video data, extracting a current observation value of a target, acquiring a previous time state vector and a covariance matrix, calculating a prior state vector at a current time, calculating an observation residual of a theoretical prediction observation value and the current observation value, dynamically updating an observation noise covariance matrix by using the observation residual and a preset forgetting factor, calculating an adaptive gain matrix based on the updated observation noise covariance matrix, correcting the prior state vector to obtain a target state estimation value at the current time, analyzing the target state estimation value into a spatial coordinate, and combining a door access terminal coordinate to generate a pan-tilt control instruction to perform linkage. The application adjusts noise parameters and gain matrices on line, overcomes the tracking lag and divergence problems caused by a fixed parameter model, and improves the reliability of video tracking in a door access linkage scene.

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology, specifically to a method and system for linking access control with video surveillance. Background Technology

[0002] The integration of access control systems and video surveillance systems is a common application model in security management. When handling abnormal access control events, the management system typically needs to activate the associated video surveillance equipment and use video analytics to dynamically track abnormal targets in the footage. This allows the system to calculate the target's spatial position and control the camera's pan-tilt-zoom (PTZ) to follow the target's movement, thereby acquiring continuous monitoring footage.

[0003] In existing video-linked tracking schemes, to estimate the continuous spatial position of a target, the system typically employs a state estimation-based filtering algorithm for trajectory extrapolation. These methods often use a fixed noise covariance parameter model, meaning that static error covariance and observation noise covariance are set during system initialization. During continuous system operation, the algorithm performs routine recursive calculations on the target observations extracted frame by frame based on the preset fixed covariance matrix to obtain the gain matrix for subsequent time steps and output the estimated target state.

[0004] However, in real-world surveillance scenarios, the target's trajectory often undergoes dynamic changes, and sudden changes in lighting or object occlusion can cause dynamic fluctuations in the actual observation error of video measurements. When the measurement environment changes, the fixed observation noise covariance matrix in existing methods cannot reflect the true error level at the current moment, leading to computational biases in the weight allocation of the system gain matrix. This parameter fixation mechanism prevents the algorithm from using the current actual measurement state to correct the model online, causing the estimated target state to deviate from the target's true trajectory, resulting in tracking lag and computational divergence. Since the state estimation data is directly used as the spatial control benchmark for hardware linkage, the aforementioned data biases ultimately lead to errors in the pan-tilt rotation commands sent to the front-end device, causing the target to leave the video surveillance field of view and compromising reliable tracking in access control linkage scenarios. Summary of the Invention

[0005] The technical problem solved by this invention is that in the target tracking process of existing access control video surveillance linkage systems, due to the variable motion state of the target and its susceptibility to environmental interference, traditional filtering tracking methods based on fixed covariance parameters cannot adapt to changes in the dynamic characteristics of the system. When the observation error fluctuates, the fixed gain matrix will cause lag and divergence in the target state estimation, thereby causing the pan-tilt-zoom (PTZ) device to fail in linkage tracking.

[0006] To address the above problems, the present invention provides the following technical solution:

[0007] The first aspect of the present invention provides a method for linking access control with video surveillance, comprising: acquiring an access control trigger event and acquiring target video data collected by a video surveillance device associated with the access control trigger event;

[0008] Extract the current observation value of the target from the target video data; use an adaptive gain model to dynamically track and calculate the current observation value to obtain the target state estimate at the current moment;

[0009] Based on the target state estimate at the current moment, execute the linkage operation corresponding to the access control trigger event.

[0010] Preferably, the step of dynamically tracking and calculating the current observation value using an adaptive gain model to obtain the target state estimate at the current moment includes:

[0011] Obtain the system's state vector, error covariance matrix, and observation noise covariance matrix at the previous time step;

[0012] Calculate the prior state vector at the current moment based on the preset state transition equation and system control input;

[0013] Based on the error covariance matrix of the previous time step, calculate the prior error covariance matrix of the current time step.

[0014] Preferably, after calculating the prior error covariance matrix at the current moment, the method further includes: calculating the theoretical predicted observation value at the current moment based on the system's observation equation and the prior state vector at the current moment;

[0015] Calculate the observation residual between the current observation and the theoretically predicted observation at the current time;

[0016] The observation noise covariance matrix of the previous time step is updated using the observation residual to obtain the dynamic observation noise covariance matrix of the current time step.

[0017] Preferably, updating the observation noise covariance matrix of the previous time step using the observation residual to obtain the dynamic observation noise covariance matrix of the current time step includes:

[0018] Multiply the observation noise covariance matrix of the previous time step by a preset forgetting factor to obtain the first reference value; calculate the product of the observation residual and the transpose of the observation residual.

[0019] The product is weighted using the difference between the value 1 and the forgetting factor to obtain a second benchmark value;

[0020] Add the first reference value to the second reference value to obtain the dynamic observation noise covariance matrix at the current time.

[0021] Preferably, after obtaining the dynamic observation noise covariance matrix at the current moment, the method further includes:

[0022] Based on the prior error covariance matrix at the current moment, the system's observation matrix, and the dynamic observation noise covariance matrix at the current moment, calculate the adaptive gain matrix at the current moment.

[0023] The prior state vector at the current time is corrected using the observation residual and the adaptive gain matrix at the current time to obtain the target state estimate at the current time.

[0024] Preferably, the step of calculating the adaptive gain matrix at the current moment based on the prior error covariance matrix, the system's observation matrix, and the dynamic observation noise covariance matrix at the current moment includes:

[0025] Calculate the product of the observation matrix of the system, the prior error covariance matrix at the current time, and the transpose of the observation matrix of the system;

[0026] The product is added to the dynamic observation noise covariance matrix at the current time to obtain the innovation covariance matrix;

[0027] The adaptive gain matrix at the current time is obtained by multiplying the prior error covariance matrix at the current time, the transpose of the observation matrix of the system, and the inverse of the innovation covariance matrix.

[0028] Preferably, the acquisition of the access control trigger event includes:

[0029] Receive status data reported by the access control terminal, the status data including access credential verification data and door lock physical status data;

[0030] When the access credential verification data indicates verification failure or the door lock physical status data indicates abnormal opening, an access control trigger event for abnormal intrusion is generated.

[0031] Preferably, extracting the current observation value of the target from the target video data includes:

[0032] The target video data is processed by frame extraction according to a preset sampling time interval to obtain an image frame sequence;

[0033] Target detection is performed on the image frame sequence to extract the bounding box pixel coordinates of the target in the image coordinate system;

[0034] The bounding box pixel coordinates are converted into position coordinates in the physical space coordinate system, and the position coordinates are used as the current observation value.

[0035] Preferably, the step of performing the linkage operation corresponding to the access control triggering event based on the target state estimate at the current moment includes:

[0036] The target state estimate at the current moment is resolved into the target's three-dimensional spatial coordinates;

[0037] Extract the fixed spatial coordinates of the access control terminal associated with the access control trigger event;

[0038] Based on the relative positional relationship between the three-dimensional spatial coordinates and the fixed spatial coordinates, gimbal control commands are generated;

[0039] The PTZ control command is sent to the video surveillance equipment.

[0040] A second aspect of the present invention provides an access control video surveillance linkage management system, comprising:

[0041] The data acquisition module is used to acquire access control trigger events and acquire target video data collected by video surveillance equipment associated with the access control trigger events;

[0042] An observation extraction module is used to extract the current observation value of the target from the target video data;

[0043] The tracking calculation module is used to dynamically track and calculate the current observation value using an adaptive gain model to obtain the target state estimate at the current moment; the linkage execution module is used to execute the linkage operation corresponding to the access control trigger event based on the target state estimate at the current moment.

[0044] This invention provides a method and system for linking access control with video surveillance. It has the following beneficial effects:

[0045] 1. This invention utilizes observation residuals and forgetting factors to dynamically update the observation noise covariance matrix. When the target's motion state changes abruptly or the measurement environment is disturbed, causing fluctuations in observation errors, this update mechanism can adjust the covariance parameters in real time. This overcomes the defect of traditional fixed-parameter filtering models that are prone to divergence when facing nonlinear maneuvers, ensuring that the target state estimation results remain convergent.

[0046] 2. This invention calculates the innovation covariance matrix based on the dynamically updated observation noise covariance matrix, and then obtains the adaptive gain matrix at the current time. The system uses this adaptive gain matrix to correct the prior state vector, and can reasonably allocate the calculation weights of prior prediction and actual observation according to the reliability of the current observation data, thereby reducing the state estimation lag phenomenon in the dynamic tracking process of the target.

[0047] 3. This invention associates the target state estimation result with the fixed spatial coordinates of the access control physical terminal. The system extracts the pixel coordinates of the image bounding box through target detection and converts them into three-dimensional coordinates in physical space. Based on the relative positional relationship between the target state estimation coordinates and the linkage reference point, the corresponding PTZ control command is generated, realizing closed-loop control of access control abnormal triggering events and video physical space tracking. Attached Figure Description

[0048] Figure 1 This is an architecture diagram of an access control video linkage verification system according to an embodiment of the present invention;

[0049] Figure 2 This is a flowchart of the access control video linkage verification method according to an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram illustrating the spatial mapping and time delay calibration principle of an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram illustrating the principle of multimodal heterogeneous data synchronous acquisition according to an embodiment of the present invention.

[0052] Figure 5 This is a schematic diagram illustrating the integrated time delay estimation and phase shift principle of an embodiment of the present invention;

[0053] Figure 6 This is a schematic diagram illustrating the adaptive dynamic time warping verification principle of an embodiment of the present invention.

[0054] Figure 7 This is a schematic diagram illustrating the comprehensive judgment and linkage output principle of an embodiment of the present invention;

[0055] Figure 8 This is a comprehensive comparison diagram of the overall logic and experimental results of the present invention; wherein, Figure 8 (a) is the closed-loop logic diagram for single-step numerical computation. Figure 8 (b) is a comparison chart of estimated trajectories under dynamic operating conditions. Figure 8 (c) is a comparison chart of the percentage of estimation error. Detailed Implementation

[0056] The technical solutions of 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.

[0057] Example:

[0058] See attached document Figure 1 , Figure 1 This is an architecture diagram of an access control video linkage verification system according to an embodiment of the present invention.

[0059] This invention provides an access control video linkage verification system, which may include: a physical boundary control host, an IPC camera, and an edge control gateway.

[0060] The physical boundary control host is deployed at the front end of the physical defense zone. The physical boundary control host contains a microcontroller that collects analog electrical signals generated by the deformation of physical cables in real time and converts these analog signals into digital register values.

[0061] The physical boundary control host is equipped with a serial communication interface, providing a passive data access channel to the outside world. The physical boundary control host has hardware driver circuitry for controlling alarm relays.

[0062] IPC cameras are deployed around the perimeter of the physical defense zone.

[0063] The optical lens of the IPC camera covers the access control area and the corresponding physical cable boundary area.

[0064] IPC cameras have a built-in video encoding chip that outputs a compressed video stream containing prediction frames and real-time transmission control protocol status messages via Ethernet.

[0065] The edge control gateway communicates with the physical boundary control host and the IPC camera.

[0066] The edge control gateway receives the compressed video stream output by the IPC camera through the Ethernet interface and polls the register values ​​of the physical boundary control host through the serial interface.

[0067] The edge control gateway is connected in series in the hardware topology between the physical boundary control host and the link that outputs hard-linked alarm signals to the back-end security platform.

[0068] The edge control gateway is equipped with alarm input and alarm output interfaces. The alarm input interface collects the status of the alarm relay dry contacts of the physical boundary control host through an optocoupler isolation circuit. The alarm output interface reproduces or blocks the status of the alarm relay dry contacts to the backend security platform through a relay or solid-state switch.

[0069] When the edge control gateway determines that the current event is a real intrusion event, it controls the alarm output interface to be turned on to allow the hard-linked alarm signal to be released.

[0070] The edge control gateway also stores fault-safe policy parameters. These parameters define the default state of the alarm output interface when the edge control gateway loses power, experiences a processor malfunction, communication interrupts, or self-test fails.

[0071] Fail-safe policies can be set to default allow, default block, or bypass pass-through. Default allow means that when the edge control gateway malfunctions, the alarm relay signal of the physical boundary control host is directly transmitted to the backend security platform;

[0072] Default blocking means keeping the alarm output interface disconnected when the edge control gateway malfunctions; bypass pass-through means directly connecting the alarm input interface and the alarm output interface through a normally closed relay or hardware bypass switch.

[0073] In actual deployment, the corresponding fail-safe strategy can be selected according to the security level requirements;

[0074] When the edge control gateway determines that the current event is an environmental false alarm event or a non-causal event, it disconnects the control alarm output interface to cut off the hard-linked alarm signal.

[0075] The edge control gateway internally runs multi-threaded cross-computation logic.

[0076] The edge control gateway includes a calibration module, a data acquisition module, a phase shifting module, a normalization and verification module, and a comprehensive decision module.

[0077] See attached document Figure 2 , Figure 2 This is a flowchart of a access control video linkage verification method according to an embodiment of the present invention.

[0078] This invention provides a method for verifying access control video linkage, comprising the following steps:

[0079] S10, establish a spatial mapping matrix between the two-dimensional pixel coordinates of the IPC camera and the one-dimensional length topology of the physical cable, obtain the trigger test data of the set physical location, calculate the inherent asymmetric time deviation between the image encoding transmission link and the serial port polling link, and generate the heterogeneous static reference delay constant.

[0080] S20: Parse the coded block prediction information of the predicted frame from the compressed video bitstream to extract the local macroblock motion intensity sequence, simultaneously extract the physical disturbance feature sequence from the serial port message, calculate the network jitter parameter based on the RTP message timestamp and local arrival time, or parse the real-time transmission control protocol reception statistics to extract the network jitter parameter, and maintain the aligned sliding time window of the above three sets of data.

[0081] S30, when the physical boundary control host outputs an alarm trigger signal, the physical disturbance feature sequence generates a preset amplitude peak, or the local macroblock motion intensity sequence generates a preset amplitude peak, the candidate linkage event within the current sliding time window is determined, the coordinates of the geometric center macroblock where the motion change occurs are located, and the comprehensive expected time offset is calculated by combining the spatial mapping matrix, the set mechanical wave propagation constant, and the heterogeneous static reference time delay constant. The comprehensive expected time offset and its positive or negative sign are used to perform reverse phase translation compensation on the local macroblock motion intensity sequence.

[0082] S40, call the dynamic time warping algorithm to calculate the morphological correlation between the local macroblock motion intensity sequence after phase translation compensation and the physical disturbance feature sequence, and use the network jitter parameter as a dynamic penalty factor to adjust the width of the band constraint window of the dynamic time warping search matrix;

[0083] S50 performs logical judgment based on morphological correlation, spatial distribution of local macroblock motion intensity sequence, physical disturbance peak, visual motion peak and device status characteristics. When the morphological correlation is lower than the preset causal correlation threshold, the current event is judged as a non-causal correlation event.

[0084] When the spatial distribution state meets the characteristics of environmental disturbance and the morphological correlation does not reach the preset alarm threshold, the current event is judged as a natural environmental disturbance event.

[0085] When the equipment status characteristics meet the fault determination conditions, the current event is determined to be an equipment fault event;

[0086] When the morphological correlation reaches the preset alarm threshold and the peak values ​​of physical disturbance and visual motion meet the conditions for determining a real intrusion, the current event is determined to be a real intrusion event.

[0087] When the current event is determined to be a natural environmental disturbance event or a non-causal event, the alarm output interface of the control edge control gateway is disconnected to cut off the hard-linked alarm signal output by the physical boundary control host and output a false alarm blocking status.

[0088] When the current event is determined to be a real intrusion event, the alarm output interface of the control edge control gateway is turned on to allow the hardware alarm signal and trigger the linkage alarm.

[0089] The technical solution of the present invention will be described in detail below in conjunction with the above system architecture and method flow steps.

[0090] See attached document Figure 3 , Figure 3 This is a schematic diagram of spatial mapping and time delay calibration according to an embodiment of the present invention.

[0091] In the access control video linkage verification method, system initialization and heterogeneous benchmark joint calibration are performed by the calibration module inside the edge control gateway. This specifically includes establishing a spatial mapping model and deriving the heterogeneous static benchmark delay constant. This process encompasses the following sub-steps:

[0092] S101, the calibration module establishes a spatial mapping model between the two-dimensional pixel coordinates of the IPC camera and the one-dimensional length topology of the physical cables connected to the physical boundary control host.

[0093] In access control and monitoring scenarios, the video images captured by the camera are two-dimensional planar fields of view, while the force points of the electronic fence are distributed in linear space. The purpose of establishing a mapping model is to associate the local motion positions that occur in the video image with the specific distance of the actual defense zone cable, thereby providing a spatial basis for subsequent calculation of physical transmission time.

[0094] Specifically, the calibration module acquires the video output from the IPC camera and delineates the area of ​​interest based on the set access control zone boundaries.

[0095] The video frame within the region of interest consists of macroblocks arranged in a two-dimensional array.

[0096] The calibration module extracts the two-dimensional pixel coordinates of any macroblock within the region of interest. Combined with pre-stored on-site wiring topology data of the electronic fence, the two-dimensional pixel coordinates The mapping is the actual physical cable length from the physical boundary of the physical cable stress point to the sensor acquisition terminal inside the control host. .

[0097] This mapping process is achieved through a space mapping matrix function:

[0098] ;

[0099] In the formula, This represents the space mapping matrix function.

[0100] In specific implementation scenarios, the spatial mapping matrix function can be represented as a two-dimensional lookup table of discrete coordinate points, or as a continuous coordinate transformation equation based on polynomial fitting.

[0101] Through the spatial mapping matrix function, the system can convert the local positional changes captured in the visual image into absolute distance parameters on the physical cable.

[0102] For the specific construction and parameter calibration process of two-dimensional lookup tables or polynomial fitting equations, those skilled in the art can use conventional mathematical interpolation algorithms based on the coordinate correspondence of the field calibration points. The coordinate transformation principle is a well-known technology in this field and will not be elaborated here.

[0103] While constructing the spatial mapping model, the calibration module writes the mechanical wave propagation velocity constant of the electronic fence cable in the current deployment environment to the internal memory. .

[0104] Mechanical wave propagation velocity constant Determined by the material density and installation tension of the physical cable, it represents the inherent physical hysteresis property of stress waves generated by physical disturbances propagating along the one-dimensional cable to the physical boundary control host.

[0105] Based on the material properties of the alloy cables commonly used in electronic fences, this velocity constant is typically distributed within... Up to 3000 The range of values.

[0106] In actual deployment, specific values ​​can be obtained by consulting the cable's factory physical parameter table or by using an audio ranging instrument for on-site measurement.

[0107] S102, the calibration module acquires the trigger test data of the set physical location, and generates a heterogeneous static reference delay constant that characterizes the time deviation of the heterogeneous data transmission link by calculating and eliminating the pure physical wave velocity consumption.

[0108] Because the video stream is transmitted via Ethernet encoding, while the underlying voltage data is passively polled via a serial bus, the paths and protocols for reaching the gateway are different, resulting in a fixed system reception time difference.

[0109] The purpose of calculating the baseline delay constant is to isolate the differences caused by network node routing and device encoding / decoding, and to establish a time synchronization baseline between devices.

[0110] After the system is physically installed, the calibration module records transient mechanical pulse test data that occurs at a location with a known cable length.

[0111] Assume the calibration physical cable length during this test is . .

[0112] The edge control gateway continuously receives video streams from IPC cameras via the Ethernet interface and continuously polls the registers of the physical boundary control host via the serial interface.

[0113] To accurately pinpoint the timing of test events, the system pre-configures visual motion vector amplitude thresholds and underlying voltage change rate thresholds for testing. Within the set calibration test time window, the calibration module continuously compares the real-time acquired feature quantities with the corresponding thresholds.

[0114] The calibration module records the absolute timestamp corresponding to the moment when the macroblock abrupt change in the video image caused by the transient mechanical pulse, captured by the edge control gateway, exceeds the visual motion vector amplitude threshold and reaches its maximum peak value in the temporal domain. This timestamp is denoted as […]. .

[0115] Synchronously, the calibration module records the absolute timestamp corresponding to the moment when the edge control gateway, via the serial port, detects a voltage surge characteristic of the underlying sensor exceeding the underlying voltage change rate threshold and reaching the maximum peak value in the time domain. This timestamp is denoted as... .

[0116] The calibration module derives the heterogeneous static reference delay constant in reverse using the following formula. :

[0117] ;

[0118] In the formula, The term represents the time it takes for the stress wave to travel along the cable to the sensor via pure physical conduction.

[0119] The calibration module adds the difference between the video arrival timestamp and the physical acquisition timestamp to the pure physical transmission time, thereby removing the physical variables generated by the mechanical wave transmission.

[0120] The calculated heterogeneous static reference delay constant This only represents the inherent system-level asymmetric time skew between the image encoding transmission link and the serial port polling link.

[0121] The heterogeneous static reference delay constant integrates the inherent routing delay of network nodes and the encoding and decoding time of hardware devices into a clear system error reference parameter.

[0122] Finally, the calibration module stores the heterogeneous static reference delay constant in the cache as a basic compensation parameter for cross-modal data alignment during normal operation, thereby reducing the adverse effects of asynchronous transmission errors in the system during subsequent dynamic time warping verification.

[0123] See attached document Figure 4 , Figure 4 This is a schematic diagram of the principle of multimodal heterogeneous data synchronous acquisition according to an embodiment of the present invention.

[0124] In the access control video linkage verification method, the acquisition of multimodal heterogeneous data and the time base alignment operation are performed by the data acquisition module inside the edge control gateway.

[0125] This process aims to extract lightweight feature data from the underlying hardware and eliminate sampling frequency differences, specifically encompassing the following sub-steps:

[0126] S201, the data acquisition module obtains the analog voltage values ​​of the underlying continuum through the serial bus, and constructs a one-dimensional discrete physical disturbance characteristic sequence after DC component removal processing.

[0127] When subjected to external forces, the deformation of the physical cable changes the electrical state of the sensors inside the physical boundary control host, thereby generating corresponding analog electrical signals.

[0128] The data acquisition module sends data read commands to the physical boundary control host at a set fixed-frequency polling cycle and parses the returned serial port messages to obtain the raw register values.

[0129] Because of the inherent tension in the physical cable when it is stationary, the original register value contains a constant DC bias component.

[0130] The data acquisition module calculates the arithmetic mean of the raw data within a historical background sliding window with a preset long period of time, and subtracts the arithmetic mean from the raw value of the current sampling point to filter out the DC component generated by constant background tension and retain the alternating component that characterizes physical disturbance.

[0131] The time span of the long-term historical background window should be set much longer than the duration of a typical single physical intrusion action, such as 10 to 30 seconds, to avoid the intrusion signal itself interfering with the baseline average.

[0132] The data acquisition module arranges the processed discrete values ​​according to the sampling timestamp order to generate a physical disturbance feature sequence. :

[0133] ;

[0134] In the formula, Indicates in The dimensionless amplitude of the physical disturbance after filtering at any given moment. This indicates the total number of sampling points within the current sliding time window.

[0135] S202, the data acquisition module bypasses the full-size video pixel decoding process, parses the slice syntax elements, macroblock prediction information or coding tree unit prediction information of the predicted frames in the compressed video bitstream, in order to extract the local macroblock motion intensity sequence.

[0136] In mainstream video compression coding standards such as H.264 or H.265, predicted frames reduce spatial redundancy by referencing the pixel block displacements of adjacent images; this displacement parameter is the motion vector. The magnitude of the motion vector directly reflects the intensity of motion of objects within the corresponding spatial region.

[0137] To reduce the computing load on the edge control gateway, the data acquisition module skips the step of restoring the compressed bitstream to an RGB or YUV pixel matrix and does not perform full pixel reconstruction decoding. Instead, it parses the motion vector syntax information corresponding to the prediction block, macroblock, or coding tree unit in the compressed domain.

[0138] The data acquisition module locates the predicted frame in the video stream and extracts the motion vector information of all valid coded blocks or macroblocks within the region of interest defined in step S10.

[0139] The data acquisition module calculates the absolute values ​​of the magnitudes of the horizontal and vertical components of the motion vector of a single macroblock, and filters out invalid macroblocks whose absolute magnitudes are lower than a preset static ambient noise threshold, in order to eliminate background interference caused by camera sensor thermal noise on the encoded data.

[0140] The data acquisition module performs scalar summation on the absolute values ​​of the moduli of the remaining valid macroblocks within the region of interest to calculate the local macroblock motion intensity sequence corresponding to that video frame. :

[0141] ;

[0142] In the formula, Represents macroblock coordinates within the region of interest. The set, and These represent the horizontal and vertical components of the macroblock's motion vector, respectively.

[0143] The data acquisition module reassembles the motion intensity parameters according to the arrival time sequence of video frames to generate a local macroblock motion intensity sequence. :

[0144] ;

[0145] This mechanism reduces the computational overhead of the processor by transforming the spatial pixel changes of the video image into a one-dimensional time series with the same dimension as the physical perturbation feature sequence.

[0146] S203, the data acquisition module extracts network jitter parameters within the current time window based on RTP / RTCP transmission statistics.

[0147] Since video data is transmitted via Ethernet, fluctuations in network bandwidth can cause random shifts in the time interval between video frames arriving at the edge control gateway.

[0148] In the parallel thread receiving video stream data, the data acquisition module acts as the RTP video stream receiver, calculating the packet arrival interval difference based on the timestamps, sequence numbers, and local reception time of consecutive RTP packets; or parsing locally maintained RTCP reception statistics to obtain the packet arrival interval jitter value. The data acquisition module calculates the mean or weighted mean of this packet arrival interval jitter value within the current sliding time window, generating a real-time network jitter parameter. .

[0149] The network jitter parameter Used to quantify the degree of congestion and instability of the current network link.

[0150] S204, the data acquisition module uses linear interpolation to unify the physical disturbance feature sequence and the local macroblock motion intensity sequence to the same discrete time reference and sliding window length.

[0151] Because the polling frequency of the serial bus is usually inconsistent with the video frame rate of the IPC camera, the discrete sampling points of the two feature sequences on the time axis cannot correspond one-to-one.

[0152] The data acquisition module uses the local system absolute clock of the edge control gateway as a unified timestamp reference to record the arrival time of various heterogeneous data.

[0153] The data acquisition module establishes a sliding time window with a fixed time span in memory and sets a uniform discrete resampling period for the entire system. .

[0154] The data acquisition module is based on a set discrete resampling period. The obtained physical disturbance feature sequence With local macroblock motion intensity sequence Perform linear interpolation resampling to achieve effective synchronization of heterogeneous data on a time base.

[0155] For the specific calculation of interpolation coefficients and time-domain mapping algorithm of linear interpolation resampling, those skilled in the art can call standard digital signal processing library functions according to the set sampling frequency ratio. The resampling processing principle is a well-known technology in this field and will not be elaborated here. The three sets of data after reference alignment are stored in the memory buffer of the edge control gateway for subsequent phase calibration and calculation.

[0156] During routine monitoring, the edge control gateway uses the alarm trigger signal or physical disturbance feature sequence exceeding the limit event output by the physical boundary control host as the main trigger condition, and the local macroblock motion intensity sequence as the auxiliary verification condition.

[0157] When a physical alarm is triggered, the edge control gateway retrieves the video motion peak within a sliding time window that includes the time of the alarm trigger.

[0158] If a valid video motion peak is found, then spatial mapping, phase shifting, and dynamic time warping checks are performed.

[0159] If no valid video motion peak is found, the system will combine video link anomaly flags, physical disturbance peaks, and device status characteristics to determine device fault, low confidence alarm, or environmental false alarm.

[0160] This avoids situations where the system cannot complete the linkage verification due to camera obstruction, frame loss, backlighting, or abnormal network transmission.

[0161] See attached document Figure 5 , Figure 5 This is a schematic diagram of integrated time delay estimation and phase shift according to an embodiment of the present invention.

[0162] In the access control video linkage verification method, the comprehensive expected delay estimation and reverse phase shift operation are performed by the phase shift module inside the edge control gateway.

[0163] The phase translation module calculates the combined effect of the propagation time of physical waves in the medium and the inherent delay of software and hardware, extrapolates the theoretical arrival time difference of data of different modes, and coarsely aligns the heterogeneous data on the time base.

[0164] Specifically, it covers the following sub-steps:

[0165] S301, the phase translation module monitors the physical disturbance characteristic sequence, local macroblock motion intensity sequence, and alarm trigger status of the physical boundary control host output by the data acquisition module.

[0166] When the physical boundary control host outputs an alarm trigger signal, the physical disturbance feature sequence generates a preset amplitude peak, or the local macroblock motion intensity sequence generates a preset amplitude peak, the phase translation module marks the corresponding sliding time window as a candidate linkage event window, and locates the coordinates of the geometric center macroblock where the motion change occurs within the candidate linkage event window.

[0167] The phase shift module pre-sets an environmental dynamic baseline threshold. This environmental dynamic baseline threshold is not a fixed constant, but rather a floating value dynamically generated by the phase shift module based on the motion intensity sequence within a historical fixed time window and superimposed with a preset environmental bias constant. This adapts to changes in the background noise baseline caused by natural factors such as wind, rain, and snow.

[0168] When the amplitude of the local macroblock motion intensity sequence continuously exceeds the dynamic baseline threshold of the environment and a local maximum occurs, a significant motion abrupt event is determined to have occurred in the video frame.

[0169] To determine the exact location of the motion abrupt event in the frame, the phase translation module extracts the two-dimensional pixel coordinates of all valid motion macroblocks within the video frame that triggered the peak.

[0170] Among them, an effective motion macroblock refers to a macroblock whose motion vector magnitude is greater than a preset low-noise rejection threshold, thereby filtering out noise and flicker caused by the camera's night gain.

[0171] The phase translation module uses the motion vector magnitude of each effective motion macroblock as a weighting factor to calculate the weighted centroid coordinates of the motion region within the video frame, and uses these coordinates as the geometric center macroblock coordinates. .

[0172] The specific formula for calculating the centroid is as follows:

[0173] ;

[0174] ;

[0175] In the formula, The total number of valid motion macroblocks within the video frame that triggers the peak. Representing the Two-dimensional pixel coordinates of an effective motion macroblock. Representing the The motion vector magnitude of an effective motion macroblock.

[0176] This weighted centroid calculation method helps reduce the interference of minor, scattered movements at the edges of the image on the overall positioning results, and calculates the coordinates of the core area where the movement is most intense.

[0177] S302, the phase translation module combines the spatial mapping matrix, the set mechanical wave propagation constant, and the heterogeneous static reference delay constant to calculate the comprehensive expected lag time.

[0178] The time difference between the occurrence of a physical deformation event at a security site and its detection by the system is composed of the combined time of spatial physical propagation and the time of heterogeneous transmission links.

[0179] The system dynamically calculates this comprehensive time difference based on the actual physical location of the mutation in the image.

[0180] The phase translation module will calculate the coordinates of the geometric center macroblock. The data is input into the spatial mapping matrix function constructed in step 510 to obtain the distance between the actual physical cable location where the mutation event occurred and the underlying sensor, denoted as the event cable length. .

[0181] Subsequently, the phase shift module adjusts the phase shift based on the mechanical wave propagation velocity constant of the physical cable. The pure physical delay time was calculated.

[0182] The phase shift module compares the pure physical delay time with the heterogeneous static reference delay constant obtained during the system initialization phase. Combined, a comprehensive expected time offset is generated. .

[0183] The specific calculation model is as follows:

[0184] ;

[0185] In the formula, This represents the theoretical physical transmission time of the physical stress wave from the point of intrusion to the underlying sensor.

[0186] Overall expected time offset Taking into account the spatial propagation delay of mechanical waves and the heterogeneous system error between image encoding and serial port polling, the theoretical relative time offset between the physical disturbance feature sequence and the video motion feature sequence is characterized.

[0187] when When the value is positive, it indicates that the physical perturbation feature sequence lags behind the video motion feature sequence;

[0188] when When the value is negative, it indicates that the physical disturbance feature sequence is ahead of the video motion feature sequence.

[0189] S303, the phase translation module uses the comprehensive expected lag time to perform reverse phase translation compensation on the local macroblock motion intensity sequence.

[0190] Because the local macroblock motion intensity sequence and the physical perturbation feature sequence have the aforementioned hysteresis misalignment on the time axis, direct morphological comparison will lead to a baseline deviation in the correlation calculation.

[0191] The phase translation module translates the local macroblock motion intensity sequence on the time axis, so that the visual feature sequence and the physical feature sequence are aligned on the theoretical basis.

[0192] The phase shift module operates according to the set discrete resampling period. Calculate the distance step size required for translation. :

[0193] ;

[0194] In the formula, This indicates the rounding operation.

[0195] The phase translation module converts the original local macroblock motion intensity sequence Perform time-axis translation to generate a phase-translation compensated local macroblock motion intensity sequence. :

[0196] ;

[0197] when When the value is positive, the phase translation module shifts the local macroblock motion intensity sequence forward along the time axis. One sampling point;

[0198] when When the value is negative, the phase translation module shifts the local macroblock motion intensity sequence backward along the time axis. A number of sampling points are used to align video motion features with physical perturbation features on a theoretical time reference.

[0199] For pointer offset or circular shift operations of discrete sequences in memory arrays, those skilled in the art can use conventional software data structure processing methods to implement them. The data shift principle is a well-known technology in this field and will not be elaborated here.

[0200] See attached document Figure 6 , Figure 6 This is a schematic diagram of the adaptive dynamic time warping verification principle according to an embodiment of the present invention.

[0201] In the access control video linkage verification method, the sequence morphology consistency comparison operation is performed by the regularization verification module inside the edge control gateway.

[0202] Because heterogeneous sensors inevitably exhibit local time axis distortion and nonlinear stretching during data acquisition, simple linear comparisons cannot truly reflect the correlation between the two.

[0203] The regularization verification module quantifies the intrinsic causal correlation between visual motion features and physical stress features using a nonlinear comparison algorithm with adaptive constraints, specifically encompassing the following sub-steps:

[0204] S401, the regularization verification module extracts the network jitter parameter within the current time window and uses this network jitter parameter as a dynamic penalty factor to calculate the width of the band constraint window of the dynamic time regularization search matrix.

[0205] In real-world security network environments, video frames experience random jitter upon arrival.

[0206] The phase shift in step S30 can only achieve coarse alignment of the time axis based on the theoretical mean.

[0207] When network jitter intensifies, the relative temporal relationships within the visual feature sequence will be locally distorted, requiring an expansion of the morphological matching search range.

[0208] Conversely, when network transmission is stable, the search range should be narrowed to prevent unrelated environmental noise from being forcibly matched.

[0209] The regularization verification module uses a strategy of linear mapping combined with boundary truncation to calculate the half-width of the band constraint window. :

[0210] ;

[0211] In the formula, This represents the real-time network jitter parameter extracted by the data acquisition module, usually measured in milliseconds.

[0212] Represents the discrete resampling period;

[0213] Used to convert network time jitter into the number of discrete sampling points;

[0214] This represents the preset jitter sensitivity amplification factor, which generally ranges from 0.5 to 2.0;

[0215] This represents the basic fault-tolerant window constant under ideal conditions without network jitter, and its value is typically taken as 1 to 3 sampling points;

[0216] These represent the lower and upper truncation values ​​for the half-width of the window, respectively. For example, when the sequence length is 100, these values ​​can be set... .

[0217] S402, within the generated banded constraint window boundary, the warping verification module calls the dynamic time warping algorithm to calculate the optimal cumulative warping distance between the local macroblock motion intensity sequence after phase translation compensation and the physical disturbance feature sequence.

[0218] The regularization and verification module records the local macroblock motion intensity sequence after phase translation compensation as a sequence. Let its discrete elements be denoted as ;

[0219] The physical disturbance characteristic sequence is denoted as sequence. Let its discrete elements be denoted as .

[0220] The total number of discrete data points in both sequences is .

[0221] Due to sequence Characterizing pixel displacement and sequence The two represent sensor voltage fluctuations, but their physical dimensions and absolute amplitudes are different. Directly calculating the difference will cause the distance matrix to fail.

[0222] Before computation, the regularization check module uses either the minimax normalization algorithm or the Z-score normalization algorithm to normalize the sequence. with sequence Dimensionless processing was performed on each.

[0223] This process maps the amplitudes of the two heterogeneous sequences to the same numerical range (e.g., the |0,1| range), thereby eliminating the difference in absolute amplitudes and retaining only the fluctuation characteristics of the temporal morphology.

[0224] After dimensionless processing, the regularization verification module is established with a size of [size missing]. The local distance matrix is ​​calculated, and the absolute difference between corresponding index pairs is calculated in the local distance matrix. .

[0225] The regularization verification module establishes a cumulative distance matrix of the same size. And execute the dynamic programming state transition equation:

[0226] ;

[0227] During state transition calculations, the regularization verification module strictly restricts the search path to within a banded constraint window, meaning it only searches when index pairs... satisfy At that time, calculate the cumulative distance of the node;

[0228] For spatial nodes that do not meet this condition, their cumulative distance is set to infinity.

[0229] To ensure the normal startup of the dynamic programming algorithm and avoid logical dead zones due to index out-of-bounds errors, the regularization and verification module initializes the boundaries of the cumulative distance matrix before iterative calculations. , and when hour ,when hour .

[0230] When the traversal calculation reaches the endpoint At that time, the regularization and verification module extracts the terminal values. It serves as the optimal cumulative normalized distance between two sequences under the current network state constraints.

[0231] The optimal cumulative regularization distance reflects the degree of morphological difference between the two sets of heterogeneous features after excluding time lag and slight deformation interference.

[0232] S403, the regularization verification module converts the optimal cumulative regularization distance into a quantified morphological correlation index through a nonlinear mapping function.

[0233] Since the optimal cumulative regularization distance varies with the fluctuation of the absolute amplitude of the input sequence, making it difficult to perform a uniform classification decision, the regularization verification module normalizes it using an exponential decay mapping model to generate the final morphological correlation.

[0234] ;

[0235] In the formula, This represents the total number of discrete data points within the current time window, used to eliminate the influence of the time window length on the total cumulative distance.

[0236] This represents a normalized scaling factor pre-calibrated through sample testing, used to adjust the correlation index's sensitivity to attenuation changes in regular distance.

[0237] During system deployment, its average value can be calculated by collecting a large number of real-world correlation data samples. value, and will Setting it to the inverse order of this average ensures that the relevance of most real-world related events falls within a numerical range that facilitates threshold determination.

[0238] Calculated morphological correlation The value ranges from 0 to 1. The closer the value is to 1, the more consistent the motion pattern of the image captured by the camera with the force pattern of the physical cable captured by the underlying sensor after eliminating the time delay misalignment factor.

[0239] For the underlying software programming steps of finding specific backtracking paths and constructing distance matrices in the dynamic time warping algorithm, those skilled in the art can refer to the well-known sequence alignment engineering implementation specifications, which will not be elaborated here.

[0240] See attached document Figure 7 , Figure 7 This is a schematic diagram of the integrated judgment and linkage output principle according to an embodiment of the present invention.

[0241] In the access control video linkage verification method, the determination of event authenticity and the output of control commands are executed by the integrated decision module inside the edge control gateway.

[0242] This invention utilizes the joint probability distribution of multi-dimensional features to determine the final action of the hardware, specifically encompassing the following sub-steps:

[0243] S501, the comprehensive decision module extracts the feature extreme values ​​within the current time window and constructs a multi-dimensional feature vector by combining the morphological correlation index.

[0244] In complex security scenarios, a single waveform distance or extreme value is insufficient to completely distinguish between physical intrusion and mechanical noise.

[0245] While extracting the motion intensity sequence of local macroblocks, the comprehensive decision module further calculates the spatial dispersion of the moving macroblocks. .

[0246] Suppose there are a total of The effective motion macroblock, the first The two-dimensional pixel coordinates of each effective motion macroblock are ( The motion vector magnitude is (mv_i), and the coordinates of the geometric center macroblock are ( Then the spatial dispersion of the moving macroblock Calculate using the following formula:

[0247] ;

[0248] in, The larger the value, the more dispersed the effective motion macroblocks are in the frame;

[0249] The smaller the value, the more concentrated the effective motion macroblocks are in a local area.

[0250] The integrated judgment module will Compared with a preset distribution threshold, when Greater than the preset dispersion threshold and morphological correlation When the alarm threshold is below the preset threshold, the system tends to classify the current event as a false alarm caused by wind, rain, tree branches, or small animals.

[0251] The comprehensive decision module will calculate the morphological correlation obtained in step S40. Physical disturbance peak Visual motion peak Environmental dynamic baseline threshold Spatial distribution of moving macroblocks Number of effective motion macroblocks Video link anomaly flag and serial communication error flags Perform data stitching.

[0252] Among them, the spatial dispersion of moving macroblocks The effective motion macroblock coordinates are obtained by calculating the weighted variance relative to the geometric center macroblock coordinates, and are used to characterize whether the motion region exhibits a large-area discrete distribution; the number of effective motion macroblocks. Used to represent the number of coded blocks involved in motion determination in a video frame;

[0253] Video link anomaly flag Used to characterize whether an IPC camera experiences abnormalities such as stream interruption, continuous frame loss, image freezing, or obstruction; serial communication anomaly flag. It is used to characterize whether the physical boundary control host has serial port timeout, long-term register saturation or data unreadable anomalies.

[0254] Because the dimensions and orders of magnitude of the above features differ, the comprehensive decision module calls the historical averages of each dimension of features, which have been pre-calculated on the training set. Compared with historical standard deviation Each of the aforementioned feature values ​​is individually Z-score standardized to eliminate the influence of different numerical units on the neural network weights, generating a dimension of... Multidimensional feature vectors:

[0255] .

[0256] S502, the comprehensive decision module inputs the multi-dimensional feature vector into the pre-trained multilayer perceptron classification model to obtain the physical state classification result of the current event.

[0257] The multilayer perceptron classification model adopts a feedforward neural network structure, which contains an input layer, a hidden layer, and an output layer.

[0258] The input layer is configured with 8 neurons, responsible for receiving data in dimension 1. Multidimensional feature vectors .

[0259] The hidden layer consists of a first fully connected layer and a second fully connected layer, with the number of nodes configured as 16 and 8 respectively. The two layers are connected by a ReLU activation function to extract the nonlinear mapping relationship between multidimensional features.

[0260] The output layer is configured with 4 neurons and normalized using the Softmax activation function.

[0261] The four probability values ​​output by the output layer correspond to four specific business physical state categories: real intrusion state, false alarm state, equipment failure state, and non-causal relationship state.

[0262] Non-causal state characteristics: Both video motion features and physical perturbation features may exhibit abnormal fluctuations within the same sliding time window, but after phase compensation, they still do not meet the morphological correlation conditions and therefore do not have a common triggering relationship.

[0263] When the maximum probability value is less than the confidence threshold, the comprehensive decision module enters a conservative decision mode. If the physical disturbance peak value... Exceeding the preset high-risk physical threshold, or the peak of visual motion If the preset high-risk visual threshold is exceeded, the integrated judgment module will allow the hardware alarm signal and mark it as a low-confidence alarm.

[0264] If neither the peak value of physical disturbance nor the peak value of visual motion exceeds the corresponding high-risk threshold, the comprehensive judgment module will classify the current state as an environmental false alarm state or a non-causal relationship state to avoid triggering invalid on-site alarms.

[0265] S503, the integrated decision module generates corresponding control commands based on the physical state classification results, and sends them out for execution through the communication interface of the edge control gateway.

[0266] When the physical state classification result determines that it is a real intrusion state, the integrated decision module generates a lockout and alarm command, controls the alarm output interface of the edge control gateway to be turned on, so as to transmit the alarm relay status of the physical boundary control host to the back-end security platform, and drives the on-site access control equipment to cut off the access permission of the entrance and exit and trigger the sound and light alarm.

[0267] When an environmental false alarm is detected, the integrated decision module generates a silent filtering instruction to keep the alarm output interface of the edge control gateway disconnected, thereby blocking the uploading of false alarms caused by the underlying hardware exceeding the limit of a single feature.

[0268] When a non-causal relationship is determined, the integrated decision module generates a non-correlated blocking command, keeps the alarm output interface of the control edge gateway disconnected, and records the current video motion characteristics, physical disturbance characteristics, and morphological correlation in the event cache for subsequent tracing.

[0269] When a device is determined to be in a faulty state, the integrated decision module encapsulates a maintenance message containing the time and location of the abnormality, uploads it to the upper-level security management platform to prompt for maintenance, and controls the alarm output interface to enter the default open state, default blocked state, or bypass direct state according to the preset fault safety policy.

[0270] To further clarify, the construction and training process of the aforementioned multilayer perceptron classification model is performed offline before the edge control gateway leaves the factory or is initially deployed.

[0271] The training steps of the model include: collecting test data on various meteorological environments, intrusion behaviors, video occlusion, video frame loss, serial communication anomalies, and sensor anomalies in real-world security scenarios; extracting the corresponding sequence features; performing standardized preprocessing; and generating multi-dimensional feature vectors as training samples.

[0272] The training samples are manually labeled and classified, and the labels are defined using the One-Hot encoding method. For example, the true label of the real intrusion state sample is set to [1,0,0,0], the true label of the environmental false alarm state sample is set to [0,1,0,0], the true label of the equipment failure state sample is set to [0,0,1,0], and the true label of the non-causal state sample is set to [0,0,0,1].

[0273] The partitioned training sample set is input into the multilayer perceptron classification model for forward inference, and the error between the output probability distribution and the true label is calculated using the cross-entropy loss function.

[0274] The Adam optimization algorithm is used to perform backpropagation of the error in order to iteratively update the weights and bias parameters of each layer in the network.

[0275] When the loss function converges on the validation set and the classification accuracy reaches the set application standard, training is terminated and the network parameters are extracted and stored in the non-volatile memory of the comprehensive decision module. The stored network parameters include the weight matrix of each fully connected layer, the bias vector, and the historical mean of each feature dimension. Historical standard deviation And the corresponding classification threshold parameters.

[0276] During system operation, the integrated decision module only calls the fixed network parameters to perform forward inference and does not train the model in real time during on-site operation.

[0277] For the specific mathematical derivation and code implementation methods of weight updates within a neural network, those skilled in the art can refer to the programming specifications of standard deep learning frameworks. The basic training and optimization principles are well-known technologies in this field and will not be elaborated here.

[0278] Figure 8 This is a comprehensive comparison diagram of the overall logic and experimental results of the present invention.

[0279] To further illustrate the specific implementation process and actual effects of the solution described in this invention, the following is combined with... Figure 8 The logic shown is explained in detail through specific numerical calculations and bench comparison experiments.

[0280] Assume this system is applied to a dynamic tracking device, and the system sampling time interval is set to...

[0281] 1. Single-step numerical calculation closed loop

[0282] Step 1: System Initialization

[0283] exist At time , set the initial state vector of the system. (Representing the master state variable and the internal deviation variable, respectively).

[0284] The initial error covariance matrix is ​​set as a diagonal matrix. Initial observation noise variance

[0285] Step 2: State Prior Prediction

[0286] exist At any given moment, the system receives external control input. According to the state transition equation of the present invention, state prediction is performed:

[0287] ;

[0288] After substituting the system parameter matrix into the calculation, the prior predicted state vector is obtained as follows: .

[0289] Simultaneously calculate the prior covariance matrix. .

[0290] Step 3: Adaptive parameter dynamic adjustment

[0291] Sensors collected Actual observation value at time .

[0292] According to the observation equation of this invention, the theoretically predicted observation values ​​are calculated. Calculate the observation residuals: .

[0293] Combination Figure 8 The adaptive adjustment logic in (a) dynamically updates the observation noise covariance matrix using the current residual. (Setting a forgetting factor) .

[0294] Step 4: Post-hoc state update

[0295] Calculate the adaptive gain matrix of the system After substituting the values, we get .

[0296] Correcting prior states using residuals:

[0297] Finally, the updated precise state principal is obtained. This result completes the computational loop for the current time step and serves as the initial value for the recursion at the next time step.

[0298] Experimental verification and effect comparison

[0299] To verify the superiority of the above adaptation scheme in real-world complex environments, the system is configured according to... Figure 8 (a) The closed-loop logic shown in the figure was fully operational, and a high dynamic variable load bench test was conducted for a duration of 8000s. The results of the operation of the present invention were compared with those of the traditional fixed parameter scheme.

[0300] Experimental Result 1: Dynamic tracking accuracy is significantly improved

[0301] like Figure 8 (b) and Figure 8 As shown in (c), during the stage of drastic change in external load:

[0302] Traditional approach: Due to fixed parameters, it cannot respond to changes in the dynamic characteristics of the system in a timely manner. Its maximum absolute error reaches 3.45%, and the root mean square error is 1.82%. The state estimation trajectory shows obvious divergence and lag.

[0303] The solution of this invention: benefits from Figure 8 (a) features a unique adaptive adjustment mechanism that ensures the state estimation trajectory always closely follows the true baseline.

[0304] Data analysis shows that the maximum absolute error of this scheme is effectively suppressed to within 1.12%, and the root mean square error is significantly reduced to 0.65%. Compared with the traditional scheme, the overall estimation accuracy is improved by approximately 64.3%.

[0305] Experimental Result 2: Strong Robustness under Initial Value Deviation Conditions

[0306] At the start of the experiment, an initial state error of 20% was artificially injected.

[0307] Traditional solutions, due to their constant gain matrix, cannot correct bias quickly. It takes about 650 seconds for the error curve to barely converge to within the allowable band of ±2%.

[0308] The present invention immediately triggers upon detecting a large initial residual. Figure 8 (a) The gain adaptive amplification stage in the logic.

[0309] like Figure 8 (c) As shown in the curve trend in the previous section, this scheme quickly eliminated the initial deviation in only about 120s and smoothly converged to the vicinity of the true reference value.

[0310] The convergence speed is nearly 5.4 times faster than that of traditional methods, which fully demonstrates that the present invention still has extremely high reliability and strong robustness under complex and harsh working conditions.

[0311] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for coordinated management of access control and video surveillance, characterized in that, include: Obtain the access control trigger event, and obtain the target video data collected by the video surveillance device associated with the access control trigger event; Extract the current observation value of the target from the target video data; The target state estimate at the current moment is obtained by dynamically tracking and calculating the current observation value using an adaptive gain model. Based on the target state estimate at the current moment, execute the linkage operation corresponding to the access control trigger event.

2. The access control video surveillance linkage management method according to claim 1, characterized in that, The step of using an adaptive gain model to dynamically track and calculate the current observation value to obtain the target state estimate at the current moment includes: obtaining the system's state vector at the previous moment, the error covariance matrix at the previous moment, and the observation noise covariance matrix at the previous moment; Calculate the prior state vector at the current moment based on the preset state transition equation and system control input; Based on the error covariance matrix of the previous time step, calculate the prior error covariance matrix of the current time step.

3. The access control video surveillance linkage management method according to claim 2, characterized in that, After calculating the prior error covariance matrix at the current time, the method further includes: calculating the theoretical predicted observation value at the current time based on the system's observation equation and the prior state vector at the current time; and calculating the observation residual between the current observation value and the theoretical predicted observation value at the current time. The observation noise covariance matrix of the previous time step is updated using the observation residual to obtain the dynamic observation noise covariance matrix of the current time step.

4. The access control video surveillance linkage management method according to claim 3, characterized in that, The step of updating the observation noise covariance matrix of the previous time step using the observation residual to obtain the dynamic observation noise covariance matrix of the current time step includes: multiplying the observation noise covariance matrix of the previous time step by a preset forgetting factor to obtain a first reference value; and calculating the product of the observation residual and the transpose of the observation residual. The product is weighted using the difference between the value 1 and the forgetting factor to obtain a second benchmark value; Add the first reference value to the second reference value to obtain the dynamic observation noise covariance matrix at the current time.

5. The access control video surveillance linkage management method according to claim 3, characterized in that, After obtaining the dynamic observation noise covariance matrix at the current moment, the method further includes: calculating the adaptive gain matrix at the current moment based on the prior error covariance matrix at the current moment, the system's observation matrix, and the dynamic observation noise covariance matrix at the current moment. The prior state vector at the current time is corrected using the observation residual and the adaptive gain matrix at the current time to obtain the target state estimate at the current time.

6. The access control video surveillance linkage management method according to claim 5, characterized in that, The step of calculating the adaptive gain matrix at the current moment based on the prior error covariance matrix at the current moment, the system's observation matrix, and the dynamic observation noise covariance matrix at the current moment includes: calculating the product of the system's observation matrix, the prior error covariance matrix at the current moment, and the transpose of the system's observation matrix. The product is added to the dynamic observation noise covariance matrix at the current time to obtain the innovation covariance matrix; The adaptive gain matrix at the current time is obtained by multiplying the prior error covariance matrix at the current time, the transpose of the observation matrix of the system, and the inverse of the innovation covariance matrix.

7. The access control video surveillance linkage management method according to claim 1, characterized in that, The acquisition of access control trigger events includes: receiving status data reported by the access control terminal, wherein the status data includes access credential verification data and door lock physical status data; When the access credential verification data indicates verification failure or the door lock physical status data indicates abnormal opening, an access control trigger event for abnormal intrusion is generated.

8. The access control video surveillance linkage management method according to claim 1, characterized in that, Extracting the current observation value of the target from the target video data includes: performing frame extraction processing on the target video data according to a preset sampling time interval to obtain an image frame sequence; Target detection is performed on the image frame sequence to extract the bounding box pixel coordinates of the target in the image coordinate system; The bounding box pixel coordinates are converted into position coordinates in the physical space coordinate system, and the position coordinates are used as the current observation value.

9. The access control video surveillance linkage management method according to claim 1, characterized in that, The step of performing a linkage operation corresponding to the access control trigger event based on the target state estimate at the current moment includes: resolving the target state estimate at the current moment into the three-dimensional spatial coordinates of the target; Extract the fixed spatial coordinates of the access control terminal associated with the access control trigger event; Based on the relative positional relationship between the three-dimensional spatial coordinates and the fixed spatial coordinates, gimbal control commands are generated; The PTZ control command is sent to the video surveillance equipment.

10. An access control video surveillance linkage management system, comprising an access control video surveillance linkage management method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire access control trigger events and acquire target video data collected by video surveillance equipment associated with the access control trigger events; An observation extraction module is used to extract the current observation value of the target from the target video data; The tracking calculation module is used to dynamically track and calculate the current observation value using an adaptive gain model to obtain the target state estimate at the current moment. The linkage execution module is used to execute linkage operations corresponding to the access control trigger event based on the target state estimate at the current time.

Images