Iris recognition and electronic fence integrated security management method and system

Abstract

The application provides a kind of based on iris recognition and electronic fence integrated security management and control method and system, it is related to security management and control and biological characteristic identification technical field, activate the distributed fence network formed by multiple electronic fence nodes, fence state perception is carried out to target object by distributed fence network, generates the fence distance change sequence of target object relative to each node;Determine the iris collection start instruction based on the fence distance change sequence, drive the iris collection module of corresponding node to capture iris texture image data;Call pre-constructed dynamic fence topology learning model, fence distance change sequence and iris texture image data are used as joint input to carry out spatial consistency analysis, generate identity fence association feature;According to the fence association feature, generate pass control instruction, instruct target pass gate to execute pass authorization operation.The application realizes the joint consistency check of spatial perception and biological characteristics, improves the real-time and accuracy of security management and control.

Description

Technical Field

[0001] This invention relates to the field of security management and biometric identification technology, and more specifically, to a security management method and system based on iris recognition and electronic fence integration. Background Technology

[0002] In existing security management scenarios, electronic fence technology and biometric recognition technology are typically deployed as two independent systems. Electronic fence systems primarily rely on sensors to detect the presence of targets within the fenced area, triggering alarms or recording events when a target approaches or crosses the fence boundary. However, they cannot verify the identity of the target, only achieving area-level presence detection, resulting in high false alarm rates and an inability to distinguish between legitimate targets and unauthorized intruders. On the other hand, while iris recognition technology boasts extremely high authentication accuracy, in practical applications it is usually deployed at fixed gate locations, triggering the recognition process only when a target actively arrives at the collection point. It lacks the ability to proactively detect the target's movement trajectory, leading to a lag in system response. Furthermore, existing technologies lack a linkage mechanism between biometric data collection and spatial fence perception. The data from both cannot be jointly analyzed, failing to verify the consistency of identity and spatial behavior during the dynamic process of a target approaching the fence. This results in a lack of multi-dimensional fusion for access control decisions, constraining the real-time performance and accuracy of overall security management. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a security management method based on the integration of iris recognition and electronic fence, the method comprising:

[0004] Activate a distributed fence network consisting of multiple electronic fence nodes. Each electronic fence node includes an iris recognition module, a fence sensing module, and a node communication module. The fence sensing module has a sensing detection range and is configured with node orientation coordinates.

[0005] The electronic fence nodes of the distributed fence network perform fence status perception processing on the target object, and generate a fence distance change sequence of the target object relative to each electronic fence node;

[0006] Based on the fence distance change sequence, an iris acquisition start command is determined, which drives the iris acquisition module of the corresponding electronic fence node to perform iris image capture processing to obtain the iris texture image data of the target object.

[0007] A pre-built dynamic fence topology learning model is invoked, and the fence distance change sequence and the iris texture image data are used as joint inputs for spatial consistency analysis to generate the identity fence association features of the target object.

[0008] Based on the identity fence association features, an access control instruction is generated for the target object, and the access control instruction is used to instruct the target access gate to perform an access authorization operation.

[0009] Furthermore, embodiments of the present invention also provide a security management system based on iris recognition and electronic fence integration, comprising:

[0010] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described security management method based on iris recognition and electronic fence integration by executing the machine-executable instructions.

[0011] Based on the above, by constructing a distributed fence network composed of multiple electronic fence nodes and deeply linking the fence sensing detection of each node with iris image acquisition, joint consistency analysis of spatial perception data and biometric data in dynamic processes is achieved. This allows the system to predict the behavioral intent of a target object as it moves by analyzing the fence distance change sequence, and accurately trigger iris acquisition accordingly, avoiding the lag problem of traditional passive acquisition at fixed points. Simultaneously, by calling a dynamic fence topology learning model to perform spatial consistency analysis on the fence distance change sequence and iris texture image data, identity fence association features are generated. This ensures that the generation of access control commands has dual verification criteria of spatial behavior legitimacy and biometric identity authenticity, significantly improving the accuracy and anti-spoofing capability of access authorization decisions. Thus, fence perception is upgraded from simple presence detection to an intelligent control method with identity association capabilities, effectively reducing the false alarm rate and improving the system's adaptability to complex scenarios. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the security management method based on the integration of iris recognition and electronic fence provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram illustrating the deployment and perception of the security management method based on the integration of iris recognition and electronic fence provided in this embodiment of the invention.

[0014] Figure 3 This is a schematic diagram of the triggering and operation of the iris acquisition module based on the integrated security management method of iris recognition and electronic fence provided in the embodiment of the present invention.

[0015] Figure 4 This is a schematic diagram of the background monitoring and on-site management of the security management method based on the integration of iris recognition and electronic fence provided in the embodiments of the present invention. Detailed Implementation

[0016] Figure 1This is a flowchart illustrating a security management method based on iris recognition and electronic fence integration, provided in one embodiment of the present invention. This embodiment uses a security management scenario deployed in a corporate R&D park as the unified application scenario. The R&D park includes a main entrance / exit, an R&D building, an experimental building, a data center, a warehousing area, and an underground parking lot. Each building entrance / exit and key internal passageway is equipped with a distributed fence network consisting of multiple electronic fence nodes. The main entrance / exit of the R&D park and the entrances to each building are equipped with target access gates.

[0017] Step S110: Activate the distributed fence network consisting of multiple electronic fence nodes. Each electronic fence node includes an iris acquisition module, a fence sensing module, and a node communication module. The fence sensing module has a sensing detection range and is configured with node orientation coordinates.

[0018] In this embodiment, after powering on, the electronic fence nodes deployed at the entrances and exits of various buildings and key internal passages within the R&D park send a node registration request to the fence network management server via a pre-configured network address. The registration request data packet includes a node identifier, node type code, and node location coordinates. The node location coordinates are two-dimensional coordinates in a Cartesian coordinate system with the lower left corner of the R&D park's floor plan as the origin. Upon receiving the registration request, the fence network management server replies with a registration confirmation, and the electronic fence node is then activated and enters the operating state. Its fence sensing module begins operation with a preset initial sensing detection frequency and a preset initial sensing detection range, while the iris recognition module enters standby mode. All activated electronic fence nodes constitute a distributed fence network. The logical topology diagram of the distributed fence network's electronic fence nodes is stored in the fence network management server's memory in the form of an adjacency list. Each node entry in the adjacency list includes a node identifier, node location coordinates, node type code, iris recognition module status, and fence sensing module status. Before deploying the electronic fence nodes, written authorization for the collection and use of iris biometric information from all target entities was obtained through enterprise entry agreements and employee handbooks. Iris texture feature data is stored in the authorized identity feature database using irreversible transformation.

[0019] Step S120: The electronic fence nodes of the distributed fence network perform fence status perception processing on the target object to generate a fence distance change sequence of the target object relative to each electronic fence node.

[0020] In this embodiment, each electronic fence node in the distributed fence network synchronously performs fence status perception processing on target objects entering its sensing and detection range.

[0021] Step S121: Each electronic fence node in the distributed fence network synchronously starts its fence sensing module and transmits a sensing signal to the sensing range of the fence sensing module at a preset sensing frequency. When the target object enters the sensing range of any electronic fence node, the fence sensing module of that electronic fence node receives the sensing reflection signal reflected from the target object, and the electronic fence node records the first time stamp of the first receiving of the sensing reflection signal.

[0022] Each electronic fence node receives a synchronization clock signal broadcast by the fence network management server through its node communication module. Each fence sensing module synchronously transmits a pulse-modulated millimeter-wave sensing signal at a preset sensing frequency, recording the transmission timestamp of the sensing signal each time it is transmitted. When a target object enters the sensing range of an electronic fence node, the fence sensing module receives the reflected signal from the target object's surface. By matching the modulation signature of the reflected signal with the sensing signal, the module determines the transmission timestamp of the corresponding sensing signal and records the moment when a valid reflected signal is first received as the initial reception timestamp.

[0023] Step S122: Based on the time phase difference between the transmission timestamp of the sensing detection signal and the reception timestamp of the sensing reflection signal, calculate the real-time sensing distance value between the target object and the node orientation coordinates of the electronic fence node, and bind the real-time sensing distance value with the corresponding timestamp to form a single-node distance measurement record.

[0024] For each reflected signal, the electronic fence node calculates the time phase difference Δt between the received timestamp of the reflected signal and the transmitted timestamp of the corresponding sensor signal. The real-time ranging value R = c × Δt / 2, where c is the speed of electromagnetic wave propagation in air. This real-time ranging value R is bound to the received timestamp of the reflected signal to form a single-node ranging record. The single-node ranging record is a key-value pair, where the key is the received timestamp and the value is the real-time ranging value.

[0025] Step S123: For each electronic fence node in the distributed fence network, the single-node distance measurement records are serialized and organized according to the timestamp order to obtain the single-node fence distance change subsequence corresponding to each electronic fence node. The single-node fence distance change subsequences of all electronic fence nodes in the distributed fence network are spatiotemporally aligned and spliced ​​according to the spatial adjacency relationship of the node orientation coordinates and the common time axis to generate a multi-node synchronized fence distance change total sequence.

[0026] For each electronic fence node, the single-node distance measurement records are arranged in ascending order by the received timestamp, forming a single-node fence distance change subsequence. This subsequence is a two-dimensional vector sequence, where each vector element is [timestamp, distance value]. All single-node fence distance change subsequences of all electronic fence nodes are aligned using a common time axis as a reference, according to the spatial adjacency of the node orientation coordinates. If the Euclidean distance between the orientation coordinates of two electronic fence nodes is less than a preset adjacency threshold, they are considered spatially adjacent. The resulting synchronized multi-node fence distance change sequence is a three-dimensional array, with the three dimensions being the time dimension, the node number dimension, and the distance measurement data dimension.

[0027] Step S124: Extract the initial sensing event spatial coordinates corresponding to the first electronic fence node in the overall sequence of fence distance changes where the target object enters the sensing detection range, and associate these initial sensing event spatial coordinates with the ranging data of subsequent nodes in the overall sequence of fence distance changes to obtain the target object's entry orientation information.

[0028] In the overall sequence of fence distance changes, the first electronic fence node with a valid distance measurement value is searched along the time axis in a positive direction. The node's azimuth coordinates are extracted as the initial sensing event spatial coordinates. The initial sensing event spatial coordinates are then associated and stored with the distance measurement data of all nodes after that time point in the overall sequence of fence distance changes in the form of a structure to obtain the target object's entry orientation information.

[0029] Step S125: Based on the initial sensing event spatial coordinates and the overlapping coverage relationship of the sensing detection range between each node, construct the fence node transfer path chain that the target object passes through when moving within the distributed fence network.

[0030] The starting node is the electronic fence node corresponding to the initial sensing event's spatial coordinates. Assuming the current node is the starting node, in the set of its spatially adjacent nodes, search for adjacent nodes whose distance measurements monotonically decrease over time in the overall fence distance change sequence. These next-hop nodes are then appended to the end of the fence node transfer path chain, and set as the new current node. This search and appending process is repeated until the distance measurements of all nodes in the overall fence distance change sequence have been path-assigned. The fence node transfer path chain is an ordered sequence of node identifiers.

[0031] Step S126: Project and decompose the real-time sensing distance values ​​corresponding to each electronic fence node in the total sequence of fence distance changes along the transfer path chain of the fence node to generate path component distance values ​​and vertical offset component distance values. Based on the continuous change trend of the path component distance values ​​and vertical offset component distance values ​​on the common time axis, construct the spatial movement deviation trajectory of the target object within the fence network.

[0032] For each time point in the overall sequence of fence distance changes, the real-time sensing distance value of each electronic fence node is projected and decomposed along the direction vector of the fence node transfer path chain. Let u be the unit vector of the direction from the current node to the next node on the fence node transfer path chain, Rpar = R × cos(α), Rperp = R × sin(α), where α is the angle between the direction corresponding to the real-time sensing distance value and u, Rpar is the path component distance value, and Rperp is the vertical offset component distance value. The path component distance value sequence and the vertical offset component distance value sequence of all time points are arranged along a common time axis to construct a spatial motion deviation trajectory. The spatial motion deviation trajectory is a two-dimensional vector sequence, and each vector element is [path component distance value, vertical offset component distance value].

[0033] Step S127: Align and associate the spatial motion deviation trajectory, the fence node transfer path chain, and the single-node fence distance change subsequence according to a common time axis to jointly constitute the fence distance change sequence of the target object relative to each electronic fence node.

[0034] The timestamp sequence of spatial motion deviation from the trajectory, the timestamp sequence of each node in the fence node transfer path chain, and the timestamp sequence of single-node fence distance change subsequence are aligned to a common time axis. These three are then packaged into a single structure, which contains an array of spatial motion deviation trajectories, a list of fence node transfer path chains, and a list of single-node fence distance change subsequences. This structure constitutes the fence distance change sequence.

[0035] Step S130: Based on the fence distance change sequence, determine the iris acquisition start command, drive the iris acquisition module of the corresponding electronic fence node to perform iris image capture processing, and obtain the iris texture image data of the target object.

[0036] Step S131: Analyze the spatial motion deviation trajectory in the fence distance change sequence, and extract the distance convergence rate sequence and vertical offset component distance measurement sequence of the spatial motion deviation trajectory on the fence node transfer path chain; calculate the satisfaction degree of the distance convergence rate sequence with respect to the preset convergence rate threshold, and the satisfaction degree of the vertical offset component distance measurement sequence with respect to the preset offset threshold.

[0037] The path component ranging value sequence is extracted from the spatial motion deviation trajectory. The first-order difference of the path component ranging value sequence is calculated along the time axis, and the difference value is divided by the time interval to obtain the distance convergence rate sequence. Each element in the distance convergence rate sequence is compared with a preset convergence rate threshold; the satisfaction level is equal to the number of elements in the distance convergence rate sequence whose values ​​are greater than the preset convergence rate threshold divided by the total length of the distance convergence rate sequence. Similarly, each element in the vertical offset component ranging value sequence is compared with a preset offset threshold; the satisfaction level is equal to the number of elements in the vertical offset component ranging value sequence whose values ​​are less than the preset offset threshold divided by the total length of the vertical offset component ranging value sequence.

[0038] Step S132: Based on the distance convergence rate satisfaction and the vertical offset component satisfaction, calculate the iris acquisition suitability assessment value of the target object relative to each electronic fence node. The iris acquisition suitability assessment value shows a gradient change as the distance convergence rate increases and the vertical offset attenuation sequence decreases.

[0039] The iris acquisition suitability assessment value Sapt = p × Sconv + q × (1 - Soffset), where Sconv is the distance convergence rate satisfaction, Soffset is the vertical offset component satisfaction, and p and q are preset positive weight coefficients with p + q = 1. The Sapt value range is [0, 1], and the larger the value, the more suitable the iris acquisition conditions are at the electronic fence node.

[0040] Step S133: When the iris acquisition suitability assessment value enters the preset iris acquisition trigger threshold range, an iris acquisition start command is generated and bound to the node identifier of the target electronic fence node corresponding to the iris acquisition suitability assessment value. The iris acquisition start command is sent to the target electronic fence node through the node communication module. The iris acquisition start command includes an iris acquisition start timestamp and acquisition parameter configuration.

[0041] When Sapt exceeds the preset iris acquisition trigger threshold, an iris acquisition start command is generated. The node identifier field of this command is assigned the node identifier of the target electronic fence node, the iris acquisition start timestamp field is assigned the current system timestamp, and the acquisition parameter configuration field includes preset values ​​for acquisition frame rate, exposure time, and near-infrared illumination intensity. The iris acquisition start command is then sent to the IP address and port of the target electronic fence node via unicast through the node communication module.

[0042] Step S134: The iris acquisition module of the target electronic fence node responds to the iris acquisition start command, starts the near-infrared illumination source and image sensor, and performs image capture on the iris region of the target object according to the acquisition frame rate and exposure parameters defined in the acquisition parameter configuration.

[0043] After receiving the iris acquisition start command, the iris acquisition module of the target electronic fence node parses the acquisition parameter configuration in the command, drives the near-infrared illumination source to the near-infrared illumination intensity specified in the command, sets the acquisition frame rate and exposure time of the image sensor to the values ​​specified in the command, starts continuous image capture, and synchronizes the exposure trigger signal of the image sensor with the near-infrared illumination source.

[0044] Step S135: During image capture, the iris acquisition module detects the position of the eye region of the target object in real time and outputs a focusing drive signal to control the optical focusing mechanism to automatically focus on the plane where the iris of the target object is located, thereby obtaining a pre-focused iris image. The pre-focused iris image is subjected to pixel-level gradient change analysis to extract the pixel distribution of the annular transition zone at the boundary between the iris region and the sclera region, and the fine focus control amount is determined based on the sharpness peak position of the pixel distribution of the annular transition zone.

[0045] The iris acquisition module utilizes a built-in lightweight human eye detection algorithm to locate the eye region of the target object in the image in real time. The optical focusing mechanism traverses the focusing positions within the initial focusing range at a preset step size, acquiring one frame at each focusing position and calculating the average pixel gradient magnitude at the boundary between the iris and sclera regions in that frame. The focusing position with the largest average gradient magnitude is selected as the pre-focusing position, and a pre-focused iris image is acquired. Circular sampling is performed on the iris-sclera boundary in the pre-focused iris image, extracting the pixel grayscale value sequence along the ring. The distribution of the gradient magnitude of the grayscale value sequence along the ring direction is calculated, and the focusing compensation amount corresponding to the peak position of the gradient magnitude distribution is taken as the fine-tuning focusing control amount.

[0046] Step S136: Adjust the optical focusing mechanism based on the fine focusing control amount, and continuously acquire multiple frames of candidate iris texture images of the target object to obtain multiple frames of candidate iris texture images. Perform iris texture artifact filtering processing on the multiple frames of candidate iris texture images. Based on the local orientation field and anisotropic smoothing calculation of the iris texture structure, remove abnormal pixel areas corresponding to lens reflection and eyelash occlusion to generate iris texture image data of the target object.

[0047] Fine-tuning of focus is achieved by superimposing a precise focus control value onto the current position of the optical focusing mechanism. A preset number of candidate iris texture images are continuously acquired at the fine-tuning focus position. For each candidate iris texture image, a local orientation field of the iris texture structure is calculated. This local orientation field is obtained by calculating the principal direction of the gradient orientation histogram within the neighborhood of each pixel. One-dimensional anisotropic smoothing filtering is applied along the principal direction of the local orientation field, while no smoothing is applied perpendicular to the principal direction. This process preserves texture details while removing abnormal pixel regions caused by lens reflections and abnormal pixel regions caused by eyelash occlusion. The multiple candidate iris texture images after filtering out abnormal pixels are then scored for image quality, and the frame with the highest quality score is selected as the iris texture image data.

[0048] Step S140: Call the pre-built dynamic fence topology learning model, take the fence distance change sequence and the iris texture image data as joint inputs for spatial consistency analysis, and generate the identity fence association features of the target object.

[0049] Step S141: Input the fence distance change sequence into the spatial topology encoding branch of the dynamic fence topology learning model, extract the topological adjacency matrix between fence nodes and the abstract representation of the temporal motion trajectory of the target object on the fence node transfer path chain, and generate the fence spatial topology embedding vector.

[0050] The spatial topology coding branch consists of a graph attention network. The input to this network is the graph structure data corresponding to the transition path chain of the fence nodes. In this data, nodes are electronic fence nodes, and node features are vectors concatenated from the node's orientation coordinates and the single-node fence distance change subsequence. Edges connect adjacent nodes in the transition path chain. The graph attention network contains two graph attention layers. Each layer calculates the attention weights of each node's neighbors. These attention weights are equal to the inner product of the concatenated source and target node features and the attention weight vector using the LeakyReLU activation function. The neighbor node features are weighted and summed according to their attention weights, then added to the node's own features. This summation is followed by a non-linear transformation to obtain the updated node features. Finally, all node features output by the graph attention network are globally averaged and pooled to obtain the fence spatial topology embedding vector.

[0051] Step S142: Input the iris texture image data into the iris texture encoding branch of the dynamic fence topology learning model, and perform normalization unpacking and texture primitive recognition processing on the iris texture through multi-layer iris feature extraction operators to obtain the iris texture embedding vector.

[0052] The iris texture encoding branch consists of a convolutional neural network (CNN) containing three convolutional layers and two fully connected layers. The first convolutional layer uses a 5×5 kernel to convolve the input iris texture image data, outputting a 32-channel feature map, followed by batch normalization and ReLU activation. The second convolutional layer uses a 3×3 kernel, outputting a 64-channel feature map, followed by batch normalization and ReLU activation. The third convolutional layer uses a 3×3 kernel, outputting a 128-channel feature map, followed by batch normalization and ReLU activation. The output of the third convolutional layer is then subjected to global average pooling to obtain a 128-dimensional feature vector. This 128-dimensional feature vector is mapped to 256 dimensions by the first fully connected layer, and then mapped to the preset iris embedding dimension by the second fully connected layer to obtain the iris texture embedding vector.

[0053] Step S143: Input the fence spatial topology embedding vector and the iris texture embedding vector into the feature cross-fusion layer of the dynamic fence topology learning model, and calculate the bidirectional interaction weights between the elements of each dimension in the fence spatial topology embedding vector and the elements of each dimension in the iris texture embedding vector.

[0054] Let the fence spatial topology embedding vector be Etopo with dimension d1, and the iris texture embedding vector be Eiris with dimension d2. The bidirectional interaction weight matrix Wcross has the shape [d1, d2], and the matrix element Wcross(i, j) = Etopo(i) × Eiris(j) / (∑i∑jEtopo(i) × Eiris(j)), where i ranges from 1 to d1, and j ranges from 1 to d2.

[0055] Step S144: Based on the bidirectional interaction weights, perform a weighted outer product operation on the fence spatial topology embedding vector and the iris texture embedding vector to generate a high-dimensional cross-fusion tensor representation, and reconstruct the high-dimensional cross-fusion tensor representation into a fusion feature map aligned with the spatial channel dimension and the iris channel dimension.

[0056] The high-dimensional cross-fusion tensor representation Fcross has a shape of [d1, d2], and Fcross(i, j) = Wcross(i, j) × Etopo(i) × Eiris(j). Fcross is channel-compressed through a 1×1 convolutional layer, compressing the shape [d1, d2] into [d1, d2 / k], where k is the compression ratio. The compressed tensor is used as the fusion feature map.

[0057] Step S145: Input the fused feature map into the multi-scale decomposition convolutional layer of the dynamic fence topology learning model, extract the joint response patterns of fence spatial topology features and iris texture features under different receptive fields in the fused feature map, generate a multi-scale joint response feature set, perform joint recalibration processing of channel attention and spatial attention on the multi-scale joint response feature set, enhance the cross-channel collaborative expression of fence spatial topology features and iris texture features, and obtain collaborative enhancement features.

[0058] The multi-scale decomposition convolutional layer contains three parallel convolutional branches with kernel sizes of 1×1, 3×3, and 5×5, respectively. The outputs of the three branches are concatenated along the channel dimension. Channel attention and spatial attention are applied sequentially to the concatenated feature map for recalibration. Channel attention generates scaling weights for each channel through global average pooling followed by a fully connected layer. Spatial attention generates spatial location weights by taking the mean and maximum values ​​along the channel dimension followed by a convolutional layer. The recalibrated feature map is then element-wise multiplied with the original concatenated feature map to obtain the collaboratively enhanced features.

[0059] Step S146: Input the collaborative enhancement features into the identity mapping subnetwork of the dynamic fence topology learning model, project the collaborative enhancement features onto the identity feature space through a fully connected mapping to generate a preliminary identity mapping representation, extract the timestamp sequence of the target object passing through each electronic fence node from the fence node transfer path chain; associate the feature vectors corresponding to the timestamp sequence in the preliminary identity mapping representation with the fence spatial location at the corresponding time point in the fence node transfer path chain to establish an association mapping table between identity feature values ​​and fence spatial locations.

[0060] The identity mapping subnetwork consists of three fully connected layers. Each fully connected layer is followed by batch normalization and ReLU activation. The last fully connected layer outputs a preliminary identity mapping representation. The arrival timestamp of each node is extracted from the fence node transfer path chain. The corresponding feature vector fragment is retrieved from the preliminary identity mapping representation according to the timestamp. The feature vector fragment is then bound to the node's orientation coordinates and written into the association mapping table.

[0061] Step S147: Calculate the continuous change trajectory of the identity feature value of the target object as it moves within the fenced network according to the association mapping table, and extract the spatial invariant component of the continuous change trajectory as a spatially consistent identity representation. Then, splice and fuse the spatially consistent identity representation and the spatial encoding of the fence node transfer path chain to generate the identity fence association feature of the target object.

[0062] Interpolation is performed on the identity feature values ​​at each spatial location in the association mapping table to obtain the continuous change trajectory of the identity feature values ​​with spatial location. Principal component analysis is performed on this continuous change trajectory, and the first principal component is extracted as the spatially invariant component to obtain the spatially consistent identity representation. The spatially consistent identity representation is concatenated with the vector encoded by the node orientation coordinate sequence of each node in the fence node transfer path chain to generate the identity fence association feature.

[0063] Step S150: Generate access control instructions for the target object based on the identity fence association features. The access control instructions are used to instruct the target access gate to perform access authorization operations.

[0064] Step S151: Obtain the gate location coordinates and gate sensing coverage area of ​​the target access gate in the distributed fence network, and determine the spatial overlap area between the gate sensing coverage area and the sensing detection range of the adjacent electronic fence node.

[0065] Query the target access gate's location coordinates and the radius of its sensing coverage area from the fence network management server. Draw a circle with the radius of the gate's sensing coverage area and the sensing range of each electronic fence node as the radius, and calculate the spatial intersection area of ​​the two circles as the spatial overlap region. The area of ​​the spatial overlap region is equal to the area of ​​the intersection of the two circles.

[0066] Step S152: Parse the spatially consistent identity representation in the identity fence association features, call the pre-stored authorized identity feature library to perform feature similarity comparison on the spatially consistent identity representation, and generate an identity matching result identifier.

[0067] The spatially consistent identity representation vector is compared with the identity registration template vector of each authorized user in the authorized identity feature library using cosine similarity calculation. The maximum similarity value is then compared with a preset matching threshold. If the similarity value is greater than the matching threshold, the identity matching result is marked as a successful match; otherwise, the match is marked as a failed match.

[0068] Step S153: When the identity matching result identifier indicates a successful match, extract the fence node transfer path chain from the identity fence association feature, calculate the predicted passage path segment from the current electronic fence node of the target object to the target access gate, and calculate the expected arrival time interval of the target object to the target access gate based on the order of the electronic fence nodes passed through by the predicted passage path segment and the physical distance between the nodes.

[0069] Extract the current electronic fence node of the target object from the fence node transfer path chain. Using this node as the starting point and the target access gate as the ending point, generate a predicted passage path segment using a shortest path search algorithm based on the spatial adjacency relationship of the electronic fence nodes. Divide the sum of the physical distances between the electronic fence nodes traversed by the predicted passage path segment by the historical average movement rate of the target object to obtain the estimated travel time. The estimated arrival time interval is [current timestamp + estimated travel time - time margin, current timestamp + estimated travel time + time margin].

[0070] Step S154: Based on the spatial overlap region and the expected arrival time interval, determine the access authorization time window of the target access gate, and generate a gate pre-activation instruction containing the identity matching result identifier, the access authorization time window and the gate's location coordinates. Send the gate pre-activation instruction to the target access gate so that the target access gate switches its gate status to a passable state within the access authorization time window.

[0071] The start time of the passage authorization time window is equal to the start time of the expected arrival time interval, and the end time is equal to the end time of the expected arrival time interval plus the passage buffer time. A gate pre-activation command is generated and sent to the control module of the target passage gate via the local area network. The target passage gate unlocks the gate blocking mechanism within the passage authorization time window, and the gate indicator light switches to green to the passage status.

[0072] Step S155: After the target access gate is switched to the passable state, real-time access sensing data at the target access gate is collected, the access trigger signal of the real-time access sensing data is monitored, the instantaneous access timestamp corresponding to the access trigger signal is extracted, and the time boundary is compared with the access authorization time window to obtain time window matching information.

[0073] The infrared access sensor of the target access gate generates a passage trigger signal when it detects a target object passing through, and records the instantaneous passage timestamp of the trigger signal. The time window matching information is equal to the comparison result of the instantaneous passage timestamp and the start and end times of the passage authorization time window.

[0074] Step S156: Based on the time window matching information and the passage completion signal returned by the passage safety sensor of the target access gate, generate a passage event closure record, bind the passage event closure record with the identity fence association feature of the target object, and generate a passage control instruction for instructing the target access gate to perform passage authorization operation.

[0075] If the instantaneous passage timestamp falls within the passage authorization time window and the passage security sensor transmits a passage completion signal, a passage event closure record is generated. The passage event closure record includes the passage time, gate identifier, and passage result. The passage event closure record is packaged and encapsulated with the identity fence association features into a passage control instruction, which is sent to the target passage gate to execute subsequent passage authorization operations.

[0076] Step S210: Poll the node status of the electronic fence nodes in the distributed fence network to obtain the acquisition window status parameters of the iris acquisition module and the sensing detection status parameters of the fence sensing module of each electronic fence node. Construct a single node status evaluation vector based on the acquisition window status parameters and the sensing detection status parameters, and aggregate the single node status evaluation vectors of all electronic fence nodes into a network status evaluation matrix.

[0077] The fence network management server sends status query commands to each electronic fence node in the distributed fence network at a preset polling cycle. Upon receiving the status query command, each electronic fence node reads the acquisition window status parameters of its iris recognition module and the sensing detection status parameters of its fence sensing module. The acquisition window status parameters include the current operating current value of the near-infrared illumination source and the current signal-to-noise ratio value of the image sensor. The sensing detection status parameters include the current transmit power value and receive sensitivity value of the fence sensing module. These four parameter values ​​are combined into a 4-dimensional single-node status evaluation vector. The single-node status evaluation vectors of all N electronic fence nodes are arranged according to their node numbers, forming a network status evaluation matrix of shape [N, 4].

[0078] Step S220: Perform spatial filtering on the network state evaluation matrix to detect target electronic fence nodes with abnormal state fluctuations, extract the abnormal state fluctuation features of the target electronic fence nodes, match and analyze the abnormal state fluctuation features with the fault feature templates in the preset node fault mode library, and determine the fault prediction type of the target electronic fence node.

[0079] The network state evaluation matrix is ​​spatially filtered along the node dimension. Spatial filtering uses the Laplace operator to calculate the deviation vector between the state evaluation vector of each node and the mean of the state evaluation vectors of its spatial neighbors. The magnitude of the deviation vector is used as the amplitude of abnormal state fluctuations. Nodes whose abnormal state fluctuation amplitude exceeds a preset abnormal threshold are marked as target electronic fence nodes. The abnormal state fluctuation feature vector of the target electronic fence node is extracted; this feature vector consists of the sign and value of each dimension of the deviation vector. The abnormal state fluctuation feature vector is then matched with each fault feature template in the node fault mode library using cosine similarity. The node fault mode library includes iris recognition module failure templates and fence sensing module failure templates; each fault feature template is represented by a feature vector. The fault type corresponding to the fault feature template with the highest cosine similarity is selected as the fault prediction type.

[0080] Step S230: When the fault prediction type indicates that the iris acquisition module has failed, extract the topological position of the target electronic fence node in the fence node transfer path chain, determine the selection range of alternative nodes, calculate the positional proximity of each candidate electronic fence node to the target electronic fence node based on the node orientation coordinates of each candidate electronic fence node, calculate the node status health based on the acquisition window status parameters, and calculate the acquisition condition optimization level based on the iris acquisition suitability assessment value; perform a weighted comprehensive analysis of the positional proximity, node status health, and acquisition condition optimization level to generate an alternative node suitability score for each candidate electronic fence node, and sort them according to the alternative node suitability score.

[0081] If the fault prediction type is iris acquisition module failure, the topological position of the target electronic fence node in the fence node transfer path chain is extracted, and the electronic fence nodes within the range of the K nodes before and after it are used as the replacement node selection range. For each candidate electronic fence node within the replacement node selection range, the position proximity is equal to 1 minus the normalized value of the Euclidean distance between the candidate electronic fence node and the target electronic fence node divided by the maximum distance of the replacement node selection range. The node health status is equal to the ratio of the signal-to-noise ratio value in the acquisition window status parameters of the candidate electronic fence node to the preset ideal signal-to-noise ratio. The acquisition condition optimization system directly adopts the iris acquisition suitability evaluation value corresponding to the node in step S132. The replacement node suitability score Snep = u × Ppos + v × Hstate + w × Capt, where Ppos is the position proximity, Hstate is the node health status, Capt is the acquisition condition optimization system, and u, v, and w are preset weight coefficients, and u + v + w = ​​1. All candidate electronic fence nodes are sorted in descending order of Snep.

[0082] Step S240: Determine alternative electronic fence nodes according to the alternative node adaptability ranking, migrate the iris acquisition task from the target electronic fence node to the alternative electronic fence node, update the logical topology diagram of the electronic fence nodes in the distributed fence network, and replace the node identifier of the target electronic fence node with the node identifier of the alternative electronic fence node.

[0083] The candidate electronic fence node with the largest Snep in the alternative node adaptability ranking is selected as the alternative electronic fence node. The node identifier field in the iris acquisition start command generated in step S133 is modified to the node identifier of the alternative electronic fence node, and the modified iris acquisition start command is sent to the alternative electronic fence node via the node communication module. In the electronic fence node logical topology diagram, the target electronic fence node identifier in the fence node transfer path chain is replaced with the alternative electronic fence node identifier.

[0084] Step S250: Recalculate the fence distance change data involving the target electronic fence node in the fence distance change sequence, reconstruct the fence distance change data based on the node orientation coordinates of the replacement electronic fence node, update the fence distance change sequence of the target object based on the reconstructed fence distance change data, and re-call the dynamic fence topology learning model to generate the updated identity fence association features.

[0085] The single-node fence distance change subsequence corresponding to the target electronic fence node in the fence distance change sequence is replaced with estimated data using the substitute electronic fence node as a reference. The estimation method is to recalculate the real-time sensing distance value using the spatial geometric relationship between the substitute electronic fence node and the target object. After updating the fence distance change sequence, step S140 is re-executed to generate the updated identity fence association features.

[0086] Step S310: When multiple target objects are detected to enter the same local area of ​​the distributed fence network at the same time, the fence distance change sequence and iris texture image data of each target object are acquired.

[0087] When the fence network management server detects that at least two target objects have at least one common electronic fence node in the fence node transfer path chain of the fence distance change sequence within the same time window, it determines that multiple target objects have entered the same local area at the same time, and retrieves the fence distance change sequence and iris texture image data of the above-mentioned target objects.

[0088] Step S320: Perform trajectory cross-analysis on the fence distance change sequences of different target objects, extract the spatial distance change curves between the fence distance change sequences of different target objects on the common time axis, perform local extremum detection processing on the spatial distance change curves, and extract the spatiotemporal points where the spatial distance between different target objects reaches a local minimum value as the intersection points of object trajectories.

[0089] For different target objects A and B, the difference between the path component distance measurement values ​​of the spatial motion deviation trajectory of A and B at each time point on the common time axis is calculated to form a spatial distance variation curve. The first derivative zero point detection is performed on the spatial distance variation curve, and the spatiotemporal point corresponding to the zero point of the first derivative with a second derivative greater than zero is taken as the intersection point of the object trajectories.

[0090] Step S330: Construct a multi-target spatial interaction topology graph based on the intersection points of the object trajectories. The multi-target spatial interaction topology graph uses the target objects as nodes and the intersection points of the object trajectories as edges. The strength of the edges is determined by the local minimum value of the spatial distance.

[0091] In the multi-target spatial interaction topology graph, each target object is a node. If there is at least one intersection point between the trajectory of two target objects, an edge is established between the two nodes. The weight of the edge is equal to 1 minus the ratio of the local minimum of the spatial distance at the intersection point of the trajectory of the target objects to the preset distance normalization constant.

[0092] Step S340: Perform community structure detection on the multi-target spatial interaction topology map, identify group object combinations that form a clustered movement pattern within the fence network, and extract the group fence distance change sequence of the group object combinations.

[0093] A community detection algorithm based on modularity optimization is used to divide the multi-object spatial interaction topology graph into communities. If the number of nodes in a community exceeds 2, the target objects in that community are marked as a group of objects. After aligning the fence distance change sequences of each target object in the group of objects along a common time axis, the average value of the spatial motion deviation trajectory of each target object at each time point is taken as the group fence distance change sequence.

[0094] Step S350: Group and bind the iris texture image data of each target object in the group object combination with the group fence distance change sequence to generate a group identity fence association feature set. Generate a group access control instruction based on the group identity fence association feature set. The group access control instruction is used to synchronously execute the access authorization operation of the target access gate for each target object in the group object combination.

[0095] The iris texture image data of each target object in the group object combination is bound to the group fence distance change sequence. Step S140 is performed on each target object to generate identity fence association features. The identity fence association features of all target objects constitute the group identity fence association feature set. A group access control instruction is generated. The access authorization time window of the group access control instruction is the union interval of the expected arrival times of all target objects. The target access gate of the group access control instruction is the gate corresponding to the common endpoint of the fence node transfer path chain of the group object combination. The target access gate remains passable within the group access authorization time window, allowing each target object in the group object combination to pass sequentially.

[0096] Step S410: Obtain the sensing and detection historical data of the fence sensing modules of each electronic fence node in the distributed fence network within a preset historical period, construct a sensing and detection historical dataset containing the global sensing status of the fence network, and extract the sensing trigger frequency distribution and sensing intensity fluctuation pattern of each electronic fence node in different time periods based on the sensing and detection historical dataset to generate fence sensing mode features.

[0097] Historical sensing data of each electronic fence node within a preset historical period is extracted from the historical database of the fence network management server. Each record includes a timestamp and a sensing trigger flag. The day is divided into several time periods, and the total number of sensing triggers for each electronic fence node within each time period is counted to obtain the sensing trigger frequency distribution. The variance of the sensing reflection signal intensity sequence within each time period is calculated to obtain the sensing intensity fluctuation pattern. The sensing trigger frequency distribution and the sensing intensity fluctuation pattern are concatenated to form a fence sensing pattern feature vector.

[0098] Step S420: Based on the fence sensing mode characteristics of each electronic fence node, the electronic fence nodes in the distributed fence network are divided into functional areas, node clusters with similar fence sensing mode characteristics are identified, and the node clusters are mapped to the physical spatial layout of the distributed fence network to generate a functional area division map of the fence network.

[0099] K-means clustering analysis was performed on the feature vectors of the fence sensing patterns of all electronic fence nodes, with the number of clusters determined by the elbow rule. Electronic fence nodes belonging to the same cluster were divided into functional areas. The node identifiers of each cluster were rendered onto the physical spatial layout map of the fence network according to the node's azimuth coordinates, with different functional areas filled with different colors, generating a functional area division map of the fence network.

[0100] Step S430: Based on the functional area division diagram of the fence network, adjust the sensing frequency and sensing range parameters of the fence sensing module of the electronic fence node in different functional areas in a differentiated manner.

[0101] For functional areas where the trigger frequency is higher than the high activity threshold, the sensing frequency of the fence sensing module is increased by multiplying the original frequency by a preset frequency boosting factor, and the sensing range is reduced by multiplying the original range by a preset range reduction factor to improve detection accuracy. For functional areas where the trigger frequency is lower than the low activity threshold, the sensing frequency of the fence sensing module is reduced by multiplying the original frequency by a preset frequency reduction factor to save energy, while keeping the sensing range unchanged.

[0102] Step S440: Based on the differentiated adjustment of the sensing frequency and sensing range parameters, reconfigure the fence status perception strategy of the electronic fence nodes in each functional area to obtain an adaptive fence perception parameter configuration scheme, and send it to each electronic fence node through the node communication module to replace the current fence sensing module operating parameters.

[0103] The adjusted sensing frequency and sensing range parameters are packaged into a fence sensing parameter configuration scheme according to the electronic fence node identifier. This scheme is then broadcast to each electronic fence node via the node communication module. Upon receiving the scheme, each electronic fence node parses it, extracts its own sensing frequency and sensing range parameters, writes them into the operating parameter register of the fence sensing module, and replaces the current operating parameters.

[0104] Step S510: Obtain the light source intensity adjustment record of the near-infrared illumination source and the focal length adjustment record of the optical focusing mechanism during the iris texture image data acquisition process, and extract the discernibility change curve of iris texture details in the iris texture image data based on the light source intensity adjustment record and the focal length adjustment record.

[0105] Extract the light source intensity adjustment record of the near-infrared illumination source and the focal length adjustment record of the optical focusing mechanism from the adjustment log recorded in step S135. Evaluate the iris texture detail discernibility of the iris texture image data; discernibility is equal to the average gradient magnitude of the iris crypt edges in the iris texture image data. Arrange the discernibility values ​​of each frame along the time axis to form a discernibility variation curve.

[0106] Step S520: When the slope of the discernibility change curve is lower than the preset texture degradation threshold, extract the iris texture frequency domain response features of the iris texture image data of the corresponding time period, input the iris texture frequency domain response features into the pre-constructed iris texture quality degradation analysis model, and generate iris texture quality degradation types. The iris texture quality degradation types include illumination uneven degradation type and focus drift degradation type.

[0107] The first derivative of the discernibility change curve is calculated. If the first derivative is lower than a preset texture degradation threshold for multiple consecutive time points, texture degradation is determined to have occurred. Two-dimensional Fourier spectra of the iris texture image data for the corresponding time period are extracted, and the average energy of each frequency band in the spectrum is extracted as the iris texture frequency domain response feature. The iris texture quality degradation analysis model is a binary classification support vector machine model. The input is the iris texture frequency domain response feature vector, and the output is either uneven illumination degradation type or focus drift degradation type. This model determines the uneven illumination degradation type based on the high proportion of low-frequency energy and the focus drift degradation type based on the rapid attenuation of high-frequency energy.

[0108] Step S530: For the aforementioned uneven illumination degradation type, generate a partitioned intensity compensation scheme for the near-infrared illumination source, wherein the partitioned intensity compensation scheme sets differentiated near-infrared illumination intensities for different sub-regions of the iris region.

[0109] If the degradation type is uneven illumination degradation, extract the average gray value of each sub-region in the iris texture image data. Using the global average gray value as a benchmark, for sub-regions with gray values ​​lower than the benchmark, increase the near-infrared illumination intensity of their corresponding illumination zones; for sub-regions with gray values ​​higher than the benchmark, decrease the near-infrared illumination intensity of their corresponding illumination zones. The adjustment amount for each illumination zone is proportional to the difference between the gray value of that zone and the global average gray value.

[0110] Step S540: For the aforementioned focus drift degradation type, generate a focus drift compensation trajectory for the optical focusing mechanism. The focus drift compensation trajectory applies a compensation control amount opposite to the drift direction to the focusing drive signal of the optical focusing mechanism.

[0111] If the degradation type is focus drift degradation, extract the focus position change sequence over time from the focus adjustment record, and perform linear trend fitting on the focus position change sequence to obtain the focus drift rate. The focus drift compensation trajectory is to superimpose a compensation amount with the opposite sign to the focus drive signal of the optical focusing mechanism at each time step. The absolute value of the compensation amount is equal to the focus drift rate multiplied by the time step.

[0112] Step S550: Feed back the partition intensity compensation scheme and the focus drift compensation trajectory to the iris acquisition module of the corresponding electronic fence node to update the acquisition parameter configuration of the iris acquisition module.

[0113] The zonal intensity compensation scheme is encoded into lighting zonal control commands, and the focus drift compensation trajectory is encoded into focus compensation commands, which are then sent to the corresponding electronic fence nodes via the node communication module. The iris acquisition module of the electronic fence node writes the zonal intensity compensation scheme into the zonal driving parameters of the near-infrared illumination source and writes the focus drift compensation trajectory into the compensation control module of the optical focusing mechanism.

[0114] For example, the method may further include: step S610: after the target object completes the passage operation of the access gate, obtain the passage event closure record of the target object and the supplementary iris texture image data at the gate during the passage process, perform texture change detection on the supplementary iris texture image data at the gate and the iris texture image data, and extract the texture micro-change features that characterize the iris texture change over time.

[0115] The supplementary iris recognition equipment deployed at the target access gate synchronously collects supplementary iris texture image data at the gate during the passage process. The supplementary iris texture image data at the gate is then compared with the iris texture image data generated in step S130 by pixel-by-pixel difference operation. A two-dimensional Fourier transform is performed on the difference image, and the energy change values ​​of each frequency band in the spectrum are extracted as texture micro-change features.

[0116] Step S620: Superimpose the texture micro-variation features onto the spatially consistent identity representation to generate a dynamic spatially consistent identity representation with time evolution. Update the identity registration template of the target object in the pre-stored authorized identity feature library based on the dynamic spatially consistent identity representation to obtain the updated identity registration template.

[0117] The texture micro-variation feature vector and the spatially consistent identity representation vector are element-wise weighted and superimposed, with the superposition weight adaptively determined by the magnitude of the texture micro-variation feature vector, to obtain a dynamic spatially consistent identity representation. The identity registration template of the target object is located in the authorized identity feature library, and the original spatially consistent identity representation in the identity registration template is replaced with the dynamic spatially consistent identity representation, while the remaining fields remain unchanged, resulting in the updated identity registration template.

[0118] Step S630: Obtain the new identity fence association features generated when the target object enters the distributed fence network again, call the updated identity registration template to match and verify the new identity fence association features, and when the matching verification is successful, use the fence distance change sequence in the new identity fence association features and the dynamic spatial consistent identity representation to perform incremental learning training on the identity mapping subnetwork of the dynamic fence topology learning model.

[0119] When the target object re-enters the area, a new identity fence association feature is generated according to step S140. The spatially consistent identity representation in the new identity fence association feature is then verified using cosine similarity matching with the updated identity registration template. After successful verification, incremental learning training is performed on the identity mapping sub-network of the dynamic fence topology learning model, using the new fence distance change sequence as input and the dynamic spatially consistent identity representation as the supervision label. The training loss function is the mean squared error loss.

[0120] Step S640: During the incremental learning training process, freeze the network parameters of the spatial topology coding branch and the iris texture coding branch in the dynamic fence topology learning model, update only the network parameters of the identity mapping sub-network, and generate the updated identity fence association features of the target object based on the dynamic fence topology learning model after incremental learning training, for use in the subsequent generation of access control instructions.

[0121] During incremental learning training, the parameters of the spatial topology coding branch and the iris texture coding branch are set to a non-updateable state, and only the weights and bias parameters of each fully connected layer in the identity mapping subnetwork are updated through backpropagation. After training, the updated dynamic fence topology learning model is used to regenerate the identity fence association features for subsequent step S150 to generate access control instructions.

[0122] Step S710: Obtain the multi-scale joint response feature set generated by the dynamic fence topology learning model during the spatial consistency analysis process, and separate the fence spatial response component and iris texture response component from the multi-scale joint response feature set. Perform spatiotemporal decoupling processing on the fence spatial response component, and perform tangential decomposition and normal decomposition on the fence spatial response component along the direction of the fence node transfer path chain to obtain the fence spatial response tangential component flow and fence spatial response normal component flow.

[0123] From the multi-scale joint response feature set generated in step S145, the first half of the channels in the channel dimension are taken as the fence spatial response component, and the second half of the channels are taken as the iris texture response component. The directional derivative of the fence spatial response component is calculated along the tangent direction of the fence node transfer path chain to obtain the tangential component flow of the fence spatial response, and the directional derivative is calculated along the normal direction to obtain the normal component flow of the fence spatial response.

[0124] Step S720: Input the tangential component flow of the fence spatial response into the fence motion intention prediction sub-network, extract the directional inertia features and rate maintenance features of the tangential component flow of the fence spatial response over continuous time steps, and generate a motion intention prediction vector. Input the normal component flow of the fence spatial response into the fence deviation intention prediction sub-network, extract the trend of the offset amplitude and the oscillation mode of the offset direction of the normal component flow of the fence spatial response relative to the transfer path chain of the fence node, and generate a deviation intention prediction vector.

[0125] The fence motion intention prediction subnetwork is a temporal convolutional network that takes the time series of the tangential component flow of the fence spatial response as input and outputs a motion intention prediction vector, which includes the predicted motion direction and predicted motion rate. The fence deviation intention prediction subnetwork is also a temporal convolutional network that takes the time series of the normal component flow of the fence spatial response as input and outputs a deviation intention prediction vector, which includes the predicted deviation direction and predicted deviation magnitude.

[0126] Step S730: Construct a fence motion intent fusion tensor based on the motion intent prediction vector and the deviation intent prediction vector, perform intent conflict resolution processing on the fence motion intent fusion tensor, and generate a fence motion intent conflict warning when the angle between the predicted motion direction pointed to by the motion intent prediction vector and the predicted offset direction pointed to by the deviation intent prediction vector exceeds a preset conflict threshold.

[0127] The fence motion intent fusion tensor is the outer product of the motion intent prediction vector and the deviation intent prediction vector, with a shape of [predicted vector dimension, predicted vector dimension]. The angle between the predicted motion direction unit vector and the predicted deviation direction unit vector is calculated. If the angle exceeds a preset conflict threshold, a fence motion intent conflict warning is generated.

[0128] Step S740: Extract the fence distance change data segment of the target object corresponding to the fence movement intention conflict warning in the fence distance change sequence, analyze the instantaneous movement orientation and instantaneous movement rate of the target object relative to each electronic fence node in the fence distance change data segment, compare the instantaneous movement orientation and instantaneous movement rate with the access permission attributes of each functional area in the fence network functional area division diagram, and generate the target object's expected intrusion area identifier.

[0129] Extract fence distance change data segments within the time window corresponding to the fence movement intention conflict warning. Calculate the instantaneous movement orientation angle and instantaneous movement speed of the target object at each electronic fence node from the fence distance change data segments. Extrapolate the instantaneous movement orientation angle along a straight line to the area boundary of the fence network functional area division map. If the extrapolated intersection point is located within a functional area that restricts access rights, then the identifier of that functional area is used as the identifier of the area the target object is expected to intrude into.

[0130] Step S750: Determine a set of reinforced fence nodes based on the identifier of the expected intrusion area of ​​the target object, wherein the set of reinforced fence nodes includes electronic fence nodes on the boundary of the expected intrusion area of ​​the target object.

[0131] In the functional area division diagram of the fence network, the node identifiers of all electronic fence nodes on the boundary of the area to be invaded by the target object are extracted to form a set of reinforced fence nodes for the area.

[0132] Step S760: Adjust the sensing detection frequency of the fence sensing module of each electronic fence node in the set of enhanced area fence nodes to the enhanced area detection frequency, and adjust the acquisition frame rate of the iris acquisition module of each electronic fence node in the set of enhanced area fence nodes to the enhanced area acquisition frame rate.

[0133] The enhanced detection frequency for the area is equal to the original sensing detection frequency multiplied by a preset enhanced frequency multiple, and the enhanced acquisition frame rate for the area is equal to the original acquisition frame rate multiplied by a preset enhanced frame rate multiple. Frequency adjustment commands are sent to each electronic fence node in the enhanced fence node set via the node communication module.

[0134] Step S770: Based on the fence movement intention conflict warning and the node configuration parameters of the adjusted area fence reinforcement node set, generate a fence intrusion prediction and interception command, and send the fence intrusion prediction and interception command to the corresponding area fence reinforcement node set.

[0135] The fence intrusion prediction and interception command includes the identifier of the target object's expected intrusion area, the node identifiers of each node in the area's reinforced fence node set, and adjusted sensing detection frequency and acquisition frame rate parameters. The fence intrusion prediction and interception command is sent to all nodes in the area's reinforced fence node set via the node communication module.

[0136] Step S810: Extract the inter-node transfer time sequence between every two adjacent electronic fence nodes on the fence node transfer path chain in the fence distance change sequence, and calculate the inter-node transfer rate sequence based on the inter-node transfer time sequence and the physical distance between adjacent electronic fence nodes. Perform deviation analysis on the inter-node transfer rate sequence and the preset traffic rate baseline template, extract the rate anomaly segments in the inter-node transfer rate sequence that deviate from the traffic rate baseline template, and determine the path segment of the fence node transfer path chain corresponding to the rate anomaly segment.

[0137] The arrival timestamp difference of the target object between every two adjacent electronic fence nodes is extracted from the fence distance change sequence to form a node transfer time sequence. The node transfer rate is equal to the physical distance between adjacent electronic fence nodes divided by the node transfer time. The preset traffic rate baseline template is the statistical average rate range of the target object under normal traffic scenarios. Continuous time segments in the node transfer rate sequence that exceed the upper and lower bounds of the traffic rate baseline template are marked as rate anomalous segments, and the fence node transfer path chain segments corresponding to the rate anomalous segments are extracted.

[0138] Step S820: Analyze the transfer rate fluctuation pattern in the rate anomaly segment and extract the change features of the vertical offset component ranging value corresponding to the path segment in the fence distance change sequence; use the transfer rate fluctuation pattern and the change features of the vertical offset component ranging value as behavioral features describing the rate anomaly segment, and together they constitute a rate anomaly behavior description set.

[0139] The transfer rate fluctuation pattern is characterized by the variance and autocorrelation function of the transfer rate sequence between nodes within the rate anomaly segment. The variation characteristics of the vertical offset component ranging value are the mean and fluctuation amplitude of the vertical offset component ranging value within the path segment. The transfer rate fluctuation pattern and the variation characteristics of the vertical offset component ranging value are concatenated to form a rate anomaly behavior description set.

[0140] Step S830: Input the rate abnormal behavior description set into the pre-built fence behavior pattern classifier, and output the fence behavior abnormal type corresponding to the rate abnormal segment. The fence behavior abnormal type includes loitering behavior type, tailing behavior type and collision behavior type.

[0141] The fence behavior pattern classifier is an ensemble classification model based on random forest. The input is a set of descriptions of abnormal rate behaviors, and the output is either a loitering behavior type, a tailing behavior type, or a collision behavior type. The model classifies loitering behavior as characterized by repeated fluctuations in the low-speed range of the transfer rate and a wide range of swings in the vertical offset component distance measurement value; tailing behavior as characterized by a high correlation between the target object's rate and the rate of the target object behind it, with the distance consistently decreasing; and collision behavior as characterized by a sharp acceleration of the transfer rate at the end and a steep peak in the vertical offset component distance measurement value.

[0142] Step S840: When the fence behavior anomaly type is the loitering behavior type, extract the set of electronic fence nodes corresponding to the path segment, and obtain the repeated iris texture image data collected by the iris acquisition module of each electronic fence node in the electronic fence node set for the target object within the corresponding time period of the path segment. Extract iris texture micro-expression features from the repeated iris texture image data, compare the iris texture micro-expression features with the preset abnormal micro-expression feature template library, and generate a micro-expression anomaly judgment label.

[0143] Repeated iris texture image data collected from all electronic fence nodes within the path segment are extracted. Iris texture micro-expression feature extraction employs a local binary pattern feature extraction method, calculating the binary pattern encoded histogram within the neighborhood of each pixel in the iris texture image data as the micro-expression feature. The micro-expression features are compared with each template in the abnormal micro-expression feature template library using chi-square distance; if the minimum distance is less than a preset distance threshold, the micro-expression anomaly judgment flag is set to true.

[0144] Step S850: When the fence behavior anomaly type is tailing behavior type, obtain the multi-object spatial distance change curve between the fence distance change sequence of the target object corresponding to the rate anomaly segment and the fence distance change sequence of another target object within the nearest distance behind it, perform adhesion detection analysis on the multi-object spatial distance change curve, and generate a tailing behavior confirmation mark when the multi-object spatial distance change curve is continuously lower than the preset safe passage distance threshold within a preset time period.

[0145] The multi-object spatial distance change curve between two target objects is calculated according to step S320. If all sampled values ​​of the multi-object spatial distance change curve within the preset time window are lower than the safe passage distance threshold, the tailgating behavior confirmation flag is set to true.

[0146] Step S860: When the fence behavior anomaly type is collision behavior type, obtain the change value of the induction reflection signal intensity recorded by the fence sensing module of the electronic fence node at the end of the path segment, and compare the change value of the induction reflection signal intensity with the physical structure tolerance parameter of the electronic fence node to generate a fence physical collision confirmation mark.

[0147] The abrupt change in the intensity of the induced reflected signal is equal to the absolute value of the difference between the intensity of the induced reflected signal before and after the timestamp at the end of the path segment. If the abrupt change in the intensity of the induced reflected signal exceeds the maximum permissible impact signal intensity value in the physical structural tolerance parameters of the electronic fence node, the physical impact confirmation flag of the fence is set to true.

[0148] Step S870: Bind the micro-expression anomaly determination identifier, the tailing behavior confirmation identifier, or the fence physical collision confirmation identifier to the corresponding rate anomaly segment and the identity fence association feature of the target object, generate a fence behavior anomaly event record, and trigger the corresponding fence security upgrade response through the fence behavior anomaly event record.

[0149] The abnormal behavior identification identifier, abnormal rate fragment, and the target object's identity fence association characteristics are bound together into a fence behavior abnormality event record. The fence security upgrade response includes: sending the fence behavior abnormality event record to the backend management platform, sending an alarm notification to the target object's mobile communication device, and switching the target access gate in the area where the target object is located to a locked state.

[0150] Step S910: Add a back projection branch to the feature cross-fusion layer of the dynamic fence topology learning model, input the fused feature map into the back projection branch for channel separation projection processing, generate fence spatial topology reconstruction projection and iris texture reconstruction projection respectively, use the fence spatial topology reconstruction projection to autoencode and reconstruct the original fence distance change sequence, generate the reconstructed fence distance change sequence, and calculate the reconstruction deviation between the reconstructed fence distance change sequence and the original fence distance change sequence as the fence topology reconstruction consistency index.

[0151] The back projection branch comprises two parallel transposed convolutional networks. The fused feature map is input into the back projection branch, and through channel separation, the first half of the fused feature map is input into the fence spatial topology reconstruction projection network, while the second half is input into the iris texture reconstruction projection network. The fence spatial topology reconstruction projection network progressively upsamples to the same dimension as the original fence distance change sequence through transposed convolutional layers, outputting the reconstructed fence distance change sequence. Ltopo = (1 / M) × Σ(Vorig - Vrecon)², where Ltopo is the fence topology reconstruction consistency index, Vorig is the vector of each time point in the original fence distance change sequence, Vrecon is the corresponding vector in the reconstructed fence distance change sequence, and M is the total number of time points.

[0152] Step S920: The original iris texture image data is regenerated pixel by pixel using the iris texture reconstruction projection to generate reconstructed iris texture image data, and the texture detail fidelity index between the reconstructed iris texture image data and the original iris texture image data is calculated.

[0153] The iris texture reconstruction projection network outputs reconstructed iris texture image data through transposed convolutional layers. The texture detail fidelity index is equal to the structural similarity index between the original iris texture image data and the reconstructed iris texture image data.

[0154] Step S930: The fence topology reconstruction consistency index and the texture detail fidelity index are fused into a model self-supervised learning signal. The parameters of the feature cross-fusion layer and the multi-scale decomposition convolution layer of the dynamic fence topology learning model are updated through the backpropagation algorithm. During the parameter update process, the update gradient of the bidirectional interaction weights in the feature cross-fusion layer is extracted, and a sparsity constraint is applied to the update gradient of the bidirectional interaction weights. Only weight connections whose absolute update gradient values ​​exceed a preset gradient threshold are retained as active weight connections.

[0155] The model's self-supervised learning signal is Lself = -Ltopo - Ltexture. During backpropagation, the gradients of Lself with respect to the parameters of the feature cross-fusion layer and the multi-scale decomposition convolutional layer are calculated and updated. A sparsity constraint is applied to the gradient of the bidirectional interaction weight matrix Wcross in the feature cross-fusion layer: if |grad(Wcross(i,j))| < a preset gradient threshold, then grad(Wcross(i,j)) is set to 0. Only weight connections whose absolute gradient values ​​exceed the preset gradient threshold are retained as active weight connections.

[0156] Step S940: Construct a post-pruning feature cross-fusion layer of the dynamic fence topology learning model based on the active weight connections, wherein the number of weight connections retained in the post-pruning feature cross-fusion layer is less than the number of weight connections before pruning.

[0157] In the feature cross-fusion layer, the elements corresponding to inactive weight connections in the Wcross matrix are set to zero, and only the elements corresponding to active weight connections are retained to construct the pruned feature cross-fusion layer.

[0158] Step S950: Extract the dimension index of the fence spatial topology embedding vector and the dimension index of the iris texture embedding vector corresponding to the active weight connection in the pruned feature cross-fusion layer, and generate an associated dimension mapping table. The associated dimension mapping table represents dimension pairs that have significant interaction between the fence spatial topology feature dimension and the iris texture feature dimension.

[0159] For each active weighted connection, record the row and column indices of its Wcross matrix. The row index corresponds to the dimension index of the fence spatial topology embedding vector, and the column index corresponds to the dimension index of the iris texture embedding vector. Write all dimension pairs into the associated dimension mapping table.

[0160] Step S960: Use the associated dimension mapping table to perform dimension filtering on the subsequently input fence distance change sequence and iris texture image data, and retain only the fence spatial topology feature dimension and iris texture feature dimension recorded in the associated dimension mapping table to participate in spatial consistency analysis processing. Generate simplified identity fence association features based on the dimension-filtered fence distance change sequence and iris texture image data, and generate simplified access control instructions according to the simplified identity fence association features.

[0161] In subsequent reasoning, only the dimensions of the row index belonging to the associated dimension mapping table record in the fence space topology embedding vector are retained, and only the dimensions of the column index belonging to the associated dimension mapping table record in the iris texture embedding vector are retained. The remaining dimensions are set to zero. Then, the subsequent steps S144 to S150 are executed to generate a simplified access control instruction.

[0162] Step S970: Record the authorization accuracy of the access authorization operation corresponding to the simplified access control instruction. When the deviation between the authorization accuracy and the authorization accuracy before pruning does not exceed the preset accuracy tolerance threshold, replace the original feature cross-fusion layer in the dynamic fence topology learning model with the feature cross-fusion layer after pruning.

[0163] The authorization accuracy rate is calculated as the proportion of correctly identified individuals allowed through the target access gate after the issuance of the streamlined access control instruction. If the authorization accuracy rate is greater than the pre-pruning authorization accuracy rate minus the preset accuracy tolerance threshold, the parameters of the pruned feature cross-fusion layer are fixed and replace the original feature cross-fusion layer parameters.

[0164] Step S1010: Obtain the transmission power of the sensing detection signal and the receiving sensitivity parameters of the sensing reflection signal of the fence sensing module of each electronic fence node in the distributed fence network, and obtain the node orientation coordinates and the signal broadcast strength of the node communication module of each electronic fence node.

[0165] The configuration parameters of each electronic fence node are retrieved from the fence network management server, including the induction detection signal transmission power Ptx and the induction reflection signal reception sensitivity Srx. Simultaneously, the node's azimuth coordinates and the signal broadcast strength of the node's communication module are obtained.

[0166] Step S1020: Based on the transmitted power of the inductive detection signal, the received sensitivity parameters of the inductive reflection signal, and the azimuth coordinates of the node, construct a detection space coverage model for the fence node. The detection space coverage model for the fence node represents the effective detection space boundary of each electronic fence node.

[0167] The detection space coverage model of the fence node is a sphere with the node azimuth coordinates of the electronic fence node as the center and the effective detection radius Req as the radius. The effective detection radius Req is derived from the radar equation: Req=(Ptx×Gtx×Grx×λ²×σ) / ((4π)³×Srx×Lsys) to the power of 1 / 4, where Gtx is the transmit antenna gain, Grx is the receive antenna gain, λ is the operating wavelength, σ is the target radar cross-section, and Lsys is the system loss.

[0168] Step S1030: Perform spatial intersection operation on the detection space coverage model of adjacent electronic fence nodes to obtain the detection overlap volume between nodes, and calculate the overlap coverage rate of the detection overlap volume between nodes to the total detection space volume of the two adjacent electronic fence nodes.

[0169] Let the radii of two adjacent electronic fence nodes A and B be ReqA and ReqB, respectively, and the distance between their centers be dAB. The volume of the intersection of the two spheres is calculated using the formula for the volume of a spherical cap. The overlap coverage rate is equal to the intersection volume divided by the sum of the volumes of the two spheres, minus the intersection volume.

[0170] Step S1040: When the overlap coverage rate is lower than the preset lower threshold of overlap coverage, mark the adjacent electronic fence node pair as a detection coverage gap node pair, and extract the spatial location and overlap coverage missing amount of the detection coverage gap node pair; when the overlap coverage rate is not lower than the preset upper threshold of overlap coverage, mark the adjacent electronic fence node pair as a detection over-coverage node pair, and extract the spatial location and overlap coverage excess amount of the detection over-coverage node pair.

[0171] The amount of missing overlap coverage equals the preset lower threshold for overlap coverage minus the actual overlap coverage. The amount of excess overlap coverage equals the actual overlap coverage minus the preset upper threshold for overlap coverage.

[0172] Step S1050: Send the node identifier, spatial location, and overlap coverage missing amount of the detection coverage gap node pair to the fence configuration adjustment interface of the distributed fence network to trigger the fence node layout adjustment instruction. The fence node layout adjustment instruction is used to instruct the deployment of supplementary electronic fence nodes between the detection coverage gap node pairs.

[0173] The fence node layout adjustment instruction includes the coordinates of the supplementary deployment location. The coordinates of the supplementary deployment location are taken as the midpoint of the line connecting the azimuth coordinates of the two nodes to the detection coverage gap node.

[0174] Step S1060: Send the node identifier, spatial location, and overlap coverage excess of the over-coverage detection node pair to the fence configuration adjustment interface of the distributed fence network to trigger the fence node parameter tuning instruction. The fence node parameter tuning instruction is used to instruct to reduce the induction detection signal transmission power of at least one electronic fence node in the over-coverage detection node pair or adjust the node azimuth coordinates.

[0175] The fence node parameter tuning command includes the target node identifier and the adjusted sensor signal transmission power value. The adjusted sensor signal transmission power value is equal to the original sensor signal transmission power multiplied by (1 - overlap coverage excess).

[0176] Step S1070: After executing the fence node layout adjustment command or the fence node parameter optimization command in the fence configuration adjustment interface of the distributed fence network, the induction detection signal transmission power and node azimuth coordinates of each electronic fence node after adjustment are obtained again, and the adjusted fence node detection space coverage model is reconstructed. Global detection hole detection is performed on the adjusted fence node detection space coverage model to identify the coordinates of uncovered areas in the adjusted distributed fence network that do not have effective detection space coverage.

[0177] Perform a spatial union operation on the spheres of the spatial coverage model of all adjusted fence nodes, and subtract the above union from the spatial coverage of the fence network to obtain the set of coordinates of the uncovered area.

[0178] Step S1080: When the coordinates of the uncovered area exceed the preset acceptable uncovered range, a fence network detection blind spot warning is generated, and the fence network detection blind spot warning is fed back to the management and scheduling center of the distributed fence network.

[0179] If the total volume of the uncovered area exceeds the preset acceptable uncovered area volume, a fence network detection blind zone warning is generated. This warning includes the boundary information and total volume value of the uncovered area's coordinates. The fence network detection blind zone warning is then sent to the management and dispatch center via the local area network.

[0180] Step S1110: Construct an iris texture feature map from the iris texture image data, extract the iris crypt distribution topology and iris pigmentation texture pattern from the iris texture image data, and generate an iris texture feature map.

[0181] Texture feature extraction based on Gabor filters is performed on iris texture image data. A multi-directional, multi-scale Gabor filter bank is used to convolve the iris texture image data, with the response amplitude of each Gabor filter kernel forming a response channel. All response channels are concatenated along the channel dimension and then fused through a 1×1 convolutional layer to output an iris texture feature map. The iris texture feature map contains spatial location information of the iris crypt distribution topology and grayscale variation information of the iris pigmentation texture pattern.

[0182] Step S1120: Extract gait rhythm features from the fence distance change sequence, analyze the step frequency pattern and stride symmetry of the distance change cycle generated when the target object moves in the fence distance change sequence, and generate fence gait rhythm features.

[0183] The path component distance measurement sequence is extracted from the spatial motion deviation trajectory of the fence distance change sequence. Autocorrelation analysis is performed on the path component distance measurement sequence, with the autocorrelation function R(τ) = Σ(Rpar(t) × Rpar(t + τ)). The τ value corresponding to the first peak of R(τ) is taken as the gait cycle duration Tgait. The stride frequency regularity is equal to 1 / Tgait. Stride symmetry is obtained by calculating the ratio of the changes in path component distance measurement values ​​within two adjacent gait cycles; a ratio close to 1 indicates good symmetry. The stride frequency regularity and stride symmetry constitute the fence gait rhythm characteristics.

[0184] Step S1130: The texture density gradient direction of different iris regions in the iris texture feature map is correlated with the longitudinal oscillation frequency of the body center of gravity corresponding to the gait frequency pattern in the fence gait rhythm feature, and a cross-modal correlation mapping relationship between iris texture and gait rhythm is constructed.

[0185] The iris region is divided into several sectors from the iris texture feature map, and the principal direction of the texture density gradient direction histogram for each sector is calculated. The longitudinal oscillation frequency of the body's center of gravity is equal to the gait frequency pattern multiplied by a preset gait center of gravity oscillation coefficient. Using canonical correlation analysis, the principal direction sequence of the texture density gradient direction of each sector is taken as the first modal variable group, and the longitudinal oscillation frequency sequence of the body's center of gravity is taken as the second modal variable group. The canonical correlation coefficient and canonical variable pairs between the two groups of variables are calculated, and a cross-modal correlation mapping relationship is constructed. The cross-modal correlation mapping relationship is a linear mapping matrix between the first modal variable group and the second modal variable group.

[0186] Step S1140: Construct an iris-gait joint biometric template based on the cross-modal association mapping relationship. The iris-gait joint biometric template includes a correlation feature vector between the iris texture feature map and the fence gait rhythm feature.

[0187] The canonical correlation coefficients in the cross-modal association mapping are arranged from largest to smallest. The top K canonical correlation coefficients and their corresponding canonical variable weight vectors are used to construct the correlation feature vector. The iris-gait joint biometric template is a structure containing iris texture feature map, fence gait rhythm feature, and correlation feature vector.

[0188] Step S1150: Retrieve the historical passage records of the target object that match the iris-gait joint biometric template from the historical passage records of the distributed fence network, and extract the gait rhythm evolution trajectory and iris texture evolution trajectory of the target object.

[0189] Using the identity-based fence association features of the target object as the retrieval criteria, the fence distance change sequence and iris texture image data for each historical passage time point of the target object are extracted from the historical passage record database. Steps S1120 and S1110 are executed for each historical time point to generate historical fence gait rhythm features and historical iris texture feature maps. The historical fence gait rhythm features are arranged in chronological order to form a gait rhythm evolution trajectory, and the historical iris texture feature maps are arranged in chronological order to form an iris texture evolution trajectory.

[0190] Step S1160: Analyze the long-term trend of the target object's gait frequency pattern and the long-term shift trend of stride symmetry based on the gait rhythm evolution trajectory, generate gait rhythm stability evaluation parameters, and analyze the gradient direction of the target object's iris pigmentation texture pattern and the degree of morphological change of the iris crypt distribution topology based on the iris texture evolution trajectory, generate iris texture stability evaluation parameters.

[0191] Linear regression was performed on the gait frequency pattern sequence in the gait rhythm evolution trajectory, and the slope of the regression line was taken as the long-term trend of gait frequency pattern. Similarly, linear regression was performed on the stride symmetry sequence, and the absolute value of the slope of the regression line was taken as the long-term shift trend of stride symmetry. The gait rhythm stability assessment parameter is equal to the weighted sum of the normalized value of the long-term trend of gait frequency pattern and the normalized value of the long-term shift trend of stride symmetry; the smaller the value, the more stable the gait rhythm. The iris texture feature maps at adjacent time points in the iris texture evolution trajectory were differencing, and the root mean square of the pixel values ​​in the differencing images was calculated as the amplitude of iris texture change. The amplitudes of iris texture change at each time point were accumulated along the time axis to obtain the iris texture stability assessment parameter.

[0192] Step S1170: Construct a joint biometric stability index based on the gait rhythm stability evaluation parameters and the iris texture stability evaluation parameters. When the joint biometric stability index is lower than a preset stability degradation threshold, generate an identity feature template degradation warning. Bind the identity feature template degradation warning to the identity fence association features of the target object and generate an identity feature template update prompt instruction. The identity feature template update prompt instruction is used to trigger the re-collection and entry process of the identity registration template of the target object. When the re-collection and entry process is triggered, the latest identity fence association features output by the dynamic fence topology learning model are preferentially called as the temporary identity registration template of the target object to maintain the access authorization capability of the target object in the distributed fence network.

[0193] The combined biometric stability index Sstab = e × Sgait + f × Siris, where Sgait is the gait rhythm stability assessment parameter, Siris is the iris texture stability assessment parameter, and e and f are preset weighting coefficients with e + f = 1. If Sstab falls below a preset stability degradation threshold, an identity feature template degradation warning is generated. An identity feature template update prompt is generated and sent to the backend management platform and the target object's mobile communication device. Before the re-collection and entry process is completed, the latest output identity fence association features from the dynamic fence topology learning model are used as a temporary identity registration template to replace the original identity registration template in the authorized identity feature library, maintaining access authorization capability.

[0194] Step S1210: The location status of the target object is continuously monitored by the fence sensing module of the electronic fence node. When the target object enters the sensing detection range of the electronic fence node, the node orientation coordinates of the electronic fence node are analyzed and it is determined whether the node orientation coordinates are within the authorized area boundary of the target object.

[0195] The authorized area boundary is read from the authorization information of the target object. The authorized area boundary is a sequence of vertex coordinates of a polygonal region. The orientation coordinates of the electronic fence nodes are determined to be inside the polygon using a ray casting method for point-to-surface containment.

[0196] Step S1220: When the node's location coordinates are within the boundary of the authorized area, an iris authentication prompt instruction is sent to the target object through the node communication module of the electronic fence node. The iris authentication prompt instruction is used to prompt the target object to perform iris authentication at the electronic fence node. When the node's location coordinates are not within the boundary of the authorized area, an out-of-area authentication alarm instruction is generated and sent to the backend management platform. At the same time, the iris acquisition module of the electronic fence node is prohibited from responding to the target object's iris authentication request.

[0197] If the node's location coordinates are within the authorized area boundary, an iris authentication prompt command is generated and sent to the target object's mobile communication device via short-range wireless communication through the node's communication module. If the node's location coordinates are not within the authorized area boundary, an out-of-area authentication alarm command is generated and sent to the backend management platform, and the iris acquisition module of the electronic fence node is set to a prohibited response state.

[0198] Step S1230: After the iris image capture process is completed at the electronic fence node within the authorized area boundary of the target object, the iris texture image data collected by the electronic fence node is compared with the identity registration template of the target object in the authorized identity feature database to generate the iris feature comparison result.

[0199] Iris texture image data is obtained according to step S130, and its iris texture embedding vector is extracted. The cosine similarity of this iris texture embedding vector with the iris texture embedding vector in the identity registration template of the target object in the authorized identity feature library is calculated. If the cosine similarity is greater than a preset comparison threshold, the iris feature comparison result is considered successful; otherwise, the comparison is considered unsuccessful.

[0200] Step S1240: When the iris feature comparison result is successful, determine the set of accessible areas within the authorized area boundary of the target object according to the target object's permission level. The permission level includes Level 1 permission, Level 2 permission, and Level 3 permission. Level 1 permission corresponds to access only public areas and designated work areas. Level 2 permission corresponds to access to Level 1 permission areas, dedicated experimental areas, and dedicated storage areas. Level 3 permission corresponds to access to all areas. When the iris feature comparison result is unsuccessful, generate an iris authentication failure alarm command, record the iris authentication failure event of the target object, the node location coordinates of the electronic fence node, and the iris authentication failure timestamp, and send the iris authentication failure alarm command to the backend management platform.

[0201] If iris recognition is successful, the system reads the target object's permission level field value and filters out the corresponding passable area set within the authorized area boundary based on the permission level. If iris recognition fails, an iris authentication failure alarm command is generated and sent to the backend management platform, while simultaneously logging the failure event.

[0202] Step S1310: When the fence sensing module of the electronic fence node in the distributed fence network detects a fence intrusion sensing event, the intrusion event type and the set of electronic fence nodes at the location of the intrusion and the intrusion timestamp corresponding to the fence intrusion sensing event are obtained. The intrusion event type includes fence climbing event type and fence destruction event type.

[0203] The method for determining fence intrusion detection events is as follows: when the fence sensing module of an electronic fence node detects a sudden increase in the vertical offset component ranging value in the fence distance change sequence, exceeding the preset intrusion determination threshold, and the path component ranging value is zero or close to zero, it is determined to be a fence climbing event; when a sudden and drastic attenuation of the induction reflection signal intensity is detected and the fence sensing module self-test reports physical deformation, it is determined to be a fence damage event. The set of node identifiers and determination timestamps of the electronic fence nodes that triggered the intrusion determination are extracted.

[0204] Step S1320: The control host of the electronic fence node triggers the audible and visual alarm response of the audible and visual alarm device according to the fence intrusion sensing event. The audible and visual alarm response includes volume output and red flashing output. At the same time, the control host sends the fence intrusion sensing event, the set of electronic fence nodes at the intrusion location, and the intrusion timestamp to the background management platform.

[0205] The control host of the electronic fence node outputs drive signals to the audible and visual alarm device via the I / O interface. The volume output decibel value and red flashing frequency of the audible and visual alarm device are both preset alarm parameter values. Simultaneously, the control host sends intrusion event information to the backend management platform via Ethernet.

[0206] Step S1330: After receiving the fence intrusion sensing event, the background management platform automatically retrieves the real-time video monitoring screen of the video monitoring device corresponding to the set of electronic fence nodes at the intrusion location, and determines the spatial coordinate range of the intrusion location based on the intrusion location.

[0207] The backend management platform queries the video surveillance equipment deployment table based on the node coordinates of each node in the set of electronic fence nodes at the location of the intrusion, and retrieves the real-time video streams from the video surveillance equipment covering the aforementioned node coordinates. The spatial coordinate range of the intrusion area is the convex hull polygon of the node coordinates of each node in the set of electronic fence nodes at the location of the intrusion.

[0208] Step S1340: Obtain the sequence of fence distance changes within a preset time period before and after the intrusion timestamp within the spatial coordinate range of the intrusion occurrence area through the distributed fence network, and determine whether there is an authorized target object within the spatial coordinate range of the intrusion occurrence area.

[0209] Centered on the timestamp of the intrusion, extract the fence distance change sequence of all electronic fence nodes within the spatial coordinate range of the intrusion area within a preset time period before and after the intrusion. Compare the identity fence association features in the above fence distance change sequence with the authorized identity feature database to determine whether there is an authorized target object.

[0210] Step S1350: When there is an authorized target object within the spatial coordinate range of the intrusion area, obtain the identity fence association feature of the target object and extract the permission level of the target object. If the permission level of the target object is lower than the area security level corresponding to the intrusion area, generate an insufficient permission cross-regional alarm command and lock all target access gates in the intrusion area.

[0211] The security level of the area where the intrusion occurred is obtained from the functional area division map of the fence network. If the permission level of the target object is lower than the area security level, an insufficient permission crossing alarm command is generated, and a gate locking command is sent to all target access gates in the area where the intrusion occurred.

[0212] Step S1360: When there is no authorized target object within the spatial coordinate range of the intrusion area, the fence intrusion sensing event is determined to be an illegal intrusion event, an illegal intrusion alarm command is generated, the full-area early warning state of the distributed fence network is activated, and security personnel are notified to go to the location of the intrusion to handle it.

[0213] The entire area warning status flag is written into the status register of each electronic fence node in the distributed fence network, and all target access gates are switched to the locked state. A notification containing the coordinates of the intrusion location is sent to the mobile communication devices of security personnel.

[0214] Step S1410: After the target object passes through the target access gate for authorization, the distributed fence network continuously acquires the fence distance change sequence of the target object, and updates the spatial movement deviation trajectory of the target object in real time according to the fence distance change sequence.

[0215] After the target object passes through the target access gate, the distributed fence network continues to obtain the fence distance change sequence in step S120 and update the spatial movement deviation trajectory in step S126.

[0216] Step S1420: Perform real-time spatial inclusion relationship determination between the spatial movement deviation trajectory of the target object and the spatial boundary of the set of passable areas of the target object. When the spatial movement deviation trajectory indicates that the target object exceeds the spatial boundary of the set of passable areas, generate an out-of-area passage event.

[0217] Convert the path component distance and vertical offset component distance values ​​at each time point in the spatial motion deviation trajectory back to Cartesian coordinates, and determine whether the coordinates are inside the boundary polygon of the passable area set. If the coordinates are outside the boundary polygon, generate a cross-region pass event.

[0218] Step S1430: Extract the cross-regional access event corresponding to the cross-regional access event, the electronic fence node identifier of the cross-regional access location, and the deviation distance value beyond the spatial boundary, and bind the cross-regional access event timestamp, the electronic fence node identifier of the cross-regional access location, and the deviation distance value to the identity fence association feature of the target object.

[0219] The deviation distance value is equal to the shortest Euclidean distance from the target object's coordinates to the boundary polygon of the passable area set. The timestamp of the cross-border occurrence, the electronic fence node identifier of the cross-border location, and the deviation distance value are added as additional attributes to the identity fence association feature structure.

[0220] Step S1440: Based on the cumulative number of cross-border passage events recorded in the identity fence association features of the target object, and the deviation distance value corresponding to each cross-border passage event, evaluate the occurrence frequency and deviation severity respectively; based on the evaluation results of the occurrence frequency and the evaluation results of the deviation severity, calculate the cumulative cross-border risk evaluation value of the target object.

[0221] The cumulative number of out-of-area passage events, Nviol, is counted within a preset statistical period. The frequency assessment value Ffreq = Nviol / Tperiod, where Tperiod is the preset statistical period length. The deviation severity assessment value Fsev = (1 / Nviol) × Σ(Di / Dmax), where Di is the deviation distance of the i-th out-of-area passage event, Dmax is the preset maximum tolerable deviation distance, and Σ represents the summation over all out-of-area passage events. The cumulative out-of-area risk assessment value Rrisk = g × Ffreq + h × Fsev, where g and h are preset weighting coefficients.

[0222] Step S1450: When the cumulative assessment value of the cross-regional risk reaches the preset cross-regional risk threshold, a temporary permission downgrade instruction is generated for the target object. The temporary permission downgrade instruction is used to temporarily reduce the permission level of the target object to a target permission level lower than the current permission level and reduce the set of accessible areas of the target object.

[0223] If Rrisk is greater than the preset out-of-area risk threshold, the target permission level in the temporary permission downgrade instruction is equal to the current permission level reduced by one level. The method for reducing the set of passable areas is to remove the area with the highest correlation to the location where the out-of-area passage event occurred from the original set of passable areas.

[0224] Step S1460: Send the temporary permission downgrade instruction to the backend management platform and all electronic fence nodes of the distributed fence network, update the permission level and passable area set of the target object, and record the permission downgrade event log.

[0225] The temporary permission downgrade command is sent to the backend management platform via the local area network. Simultaneously, it is broadcast to all electronic fence nodes in the distributed fence network via the node communication module, and each electronic fence node updates its locally stored target object permission cache table. The timestamp of the permission downgrade event, the target object identifier, the original permission level, the new permission level, and the reason for downgrade are written to the permission downgrade event log.

[0226] Based on the above embodiments, combined with Figure 2 Content, Figure 2This is a deployment and perception diagram of the security management method based on iris recognition and electronic fence integration provided by an embodiment of the present invention. It illustrates the deployment architecture of the distributed fence network and the target object perception process in the form of a top-down plan view of an actual passage or perimeter. Multiple electronic fence nodes are drawn as pillars, with probe and camera symbols at the top. They are installed on both sides of the passage according to their actual positions. The sensing range of the fence sensing module is clearly marked in a dashed fan shape in front of each node. A target object moves along a specific direction in the passage. The diagram uses a chain of dashed arrows connecting the various electronic fence nodes to represent the fence node transfer path chain along which the target object moves. A slightly wavy solid line represents the actual spatial movement deviation of the target object from its trajectory. The node coordinates are specifically marked at the electronic fence node where the target object first enters. A waveform diagram is drawn in the blank space of the diagram, with time on the horizontal axis and distance on the vertical axis. The diagram contains multiple curves of different line types, representing the waveforms of the sensing distance values ​​measured in real time from different electronic fence nodes over time, intuitively showing the composition of the fence distance change sequence.

[0227] Combination Figure 3 Content, Figure 3 This is a schematic diagram illustrating the triggering and operation of an iris acquisition module based on an integrated security management method for iris recognition and electronic fences, provided by an embodiment of the present invention. The diagram shows the working principle of the iris acquisition module at an electronic fence node in a close-up scene. A side view of a specific electronic fence node post shows its iris acquisition module aligned with an approaching target, whose face is facing the camera. The status display screen on the node post surface is divided into two sections: the left side displays two dynamic curves, labeled as the distance convergence rate sequence and the vertical offset attenuation sequence, respectively; the right side displays a dashboard with a pointer indicating the current iris acquisition suitability assessment value. When the pointer enters the green area, the screen highlights "Acquisition Start," and the green area is marked as the iris acquisition trigger threshold range. A command bubble extends from inside the node, indicating "Iris Acquisition Start Command (including acquisition parameter configuration)." A partial cross-section of the node's iris acquisition module clearly shows its internal structure: the light emitted by the annular near-infrared illumination source is represented by a light cone; the photosensitive element of the image sensor and the operating optical focusing mechanism are indicated by arrows indicating the lens's forward and backward movement. Insert a close-up image of the eye next to the node, showing the process from the captured pre-focused iris image (blurred) to the final output high-resolution iris texture image data (clear, without eyelash obstruction and reflection) after focusing and artifact filtering.

[0228] Combination Figure 4 Content, Figure 4This is a schematic diagram of the background monitoring and on-site control of the security management method based on iris recognition and electronic fence integration provided in this embodiment of the invention. The diagram includes two main views. The first view is a schematic diagram of the interface of the security center monitoring screen. The background of the interface is an electronic map of the distributed fence network. The location, orientation coordinates, and sensing detection range of each electronic fence node are clearly marked on the map. The light point representing the target object on the map is moving along a path. The spatial movement deviation trajectory (solid line) formed by its movement overlaps with a preset fence node transfer path chain (dashed line). One side of the interface displays the real-time updated iris texture image data captured by the iris acquisition module of the target object. The core area in the center of the interface displays a window called "Dynamic Fence Topology Learning Model Output", which uses a continuously evolving synthetic 3D pattern that integrates spatial trajectory lines and iris texture feature clouds to intuitively represent the identity fence association features of the current target object being generated. The information bar at the bottom of the interface displays "Current Spatial Consistent Identity Representation Matching Degree: High" and "Association Mapping Table Updated". The second view is a real-scene combined line drawing of the passage exit, showing the collaborative management and abnormal behavior scene of the access gate. In the foreground, the target access gate is located at the end of the passage. The indicator light on top of the gate is red, indicating it is locked. The gate's sensor coverage area is drawn on the ground in front of it, overlapping with the detection range of the nearest electronic fence node. The target object is approaching the gate. In the background, there is a simplified diagram of a backend server rack. A highlighted command signal line runs from the server rack to the target access gate, labeled "Gate Pre-Activation Command." After the command is issued, the indicator light on top of the gate turns green, the gate arm opens, and it switches to a passable state. Next to the gate is labeled "Access Authorization Time Window Activated." The upper right sub-diagram shows out-of-bounds behavior: within an area with a clearly defined red boundary line, the target object's movement trajectory exceeds the boundary. Next to the trajectory is a warning symbol and the text "Out-of-bounds Access Event." The lower right sub-diagram shows group analysis: in the simplified plan view, the movement trajectories of multiple target objects intersect at a single point, which is highlighted and labeled "Object Trajectory Intersection Point." This generates a simplified network diagram connected by points and lines, labeled "Multi-Target Spatial Interaction Topology."

[0229] In an exemplary embodiment, a security management system integrating iris recognition and electronic fence is provided. This system can be a terminal, server, etc., and its internal structure includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a security management method integrating iris recognition and electronic fence. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of the iris recognition and electronic fence integrated security management system, or an external keyboard, touchpad, or mouse, etc.

[0230] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A security management method based on iris recognition and electronic fence integration, characterized in that, The method includes: Activate a distributed fence network consisting of multiple electronic fence nodes. Each electronic fence node includes an iris recognition module, a fence sensing module, and a node communication module. The fence sensing module has a sensing detection range and is configured with node orientation coordinates. The electronic fence nodes of the distributed fence network perform fence status perception processing on the target object, and generate a fence distance change sequence of the target object relative to each electronic fence node; Based on the fence distance change sequence, an iris acquisition start command is determined, which drives the iris acquisition module of the corresponding electronic fence node to perform iris image capture processing to obtain the iris texture image data of the target object. A pre-built dynamic fence topology learning model is invoked, and the fence distance change sequence and the iris texture image data are used as joint inputs for spatial consistency analysis to generate the identity fence association features of the target object. Based on the identity fence association features, an access control instruction is generated for the target object, and the access control instruction is used to instruct the target access gate to perform an access authorization operation.

2. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The step of performing fence status perception processing on the target object through the electronic fence nodes of the distributed fence network, and generating a fence distance change sequence of the target object relative to each electronic fence node, includes: Each electronic fence node in the distributed fence network synchronously activates its fence sensing module and transmits a sensing signal to the sensing range of the fence sensing module at a preset sensing frequency. When the target object enters the sensing range of any electronic fence node, the fence sensing module of that electronic fence node receives the sensing reflection signal reflected from the target object, and the electronic fence node records the first time stamp of the first receiving of the sensing reflection signal. Based on the time phase difference between the transmission timestamp of the sensing detection signal and the reception timestamp of the sensing reflection signal, the real-time sensing distance value between the target object and the node orientation coordinates of the electronic fence node is calculated, and the real-time sensing distance value is bound to the corresponding timestamp to form a single-node distance measurement record. For each electronic fence node in the distributed fence network, the single-node distance measurement records are serialized and organized according to the timestamp order to obtain the single-node fence distance change subsequence corresponding to each electronic fence node. The single-node fence distance change subsequences of all electronic fence nodes in the distributed fence network are spatiotemporally aligned and spliced ​​according to the spatial adjacency relationship of the node orientation coordinates and the common time axis to generate a multi-node synchronized fence distance change total sequence. Extract the initial sensing event spatial coordinates corresponding to the first electronic fence node in the overall sequence of fence distance changes when the target object enters the sensing detection range, and associate these initial sensing event spatial coordinates with the ranging data of subsequent nodes in the overall sequence of fence distance changes to obtain the target object's entry orientation information; Based on the spatial coordinates of the initial sensing event and the overlapping coverage relationship of the sensing detection range between each node, a fence node transfer path chain is constructed as the target object moves within the distributed fence network. The real-time sensing distance values ​​corresponding to each electronic fence node in the total sequence of fence distance changes are projected and decomposed along the transfer path chain of the fence node to generate path component distance values ​​and vertical offset component distance values. Based on the continuous change trend of the path component distance values ​​and vertical offset values ​​on the common time axis, the spatial movement deviation trajectory of the target object in the fence network is constructed. The spatial motion deviation trajectory, the fence node transfer path chain, and the single-node fence distance change subsequence are aligned and associated according to a common time axis to jointly constitute the fence distance change sequence of the target object relative to each electronic fence node.

3. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The step of determining the iris acquisition start command based on the fence distance change sequence, driving the iris acquisition module of the corresponding electronic fence node to perform iris image capture processing, and obtaining the iris texture image data of the target object includes: Analyze the spatial motion deviation trajectory in the fence distance change sequence, and extract the distance convergence rate sequence and vertical offset component distance measurement sequence on the fence node transfer path chain of the spatial motion deviation trajectory; calculate the satisfaction degree of the distance convergence rate sequence with respect to the preset convergence rate threshold, and the satisfaction degree of the vertical offset component distance measurement sequence with respect to the preset offset threshold, respectively. Based on the distance convergence rate satisfaction and the vertical offset component satisfaction, the iris acquisition suitability evaluation value of the target object relative to each electronic fence node is calculated. The iris acquisition suitability evaluation value shows a gradient change as the distance convergence rate increases and the vertical offset decay sequence decreases. When the iris acquisition suitability assessment value enters the preset iris acquisition trigger threshold range, an iris acquisition start command is generated and bound to the node identifier of the target electronic fence node corresponding to the iris acquisition suitability assessment value. The iris acquisition start command is sent to the target electronic fence node through the node communication module. The iris acquisition start command includes an iris acquisition start timestamp and acquisition parameter configuration. The iris acquisition module of the target electronic fence node responds to the iris acquisition start command, activates the near-infrared illumination source and image sensor, and performs image capture on the iris region of the target object according to the acquisition frame rate and exposure parameters specified in the acquisition parameter configuration; During image capture, the iris acquisition module detects the position of the eye region of the target object in real time and outputs a focusing drive signal to control the optical focusing mechanism to automatically focus on the plane where the iris of the target object is located, thereby obtaining a pre-focused iris image. The pre-focused iris image is subjected to pixel-level gradient change analysis to extract the pixel distribution of the annular transition zone at the boundary between the iris region and the sclera region, and the fine focus control amount is determined based on the sharpness peak position of the pixel distribution of the annular transition zone. Based on the fine focus control, the optical focusing mechanism is adjusted, and multiple consecutive acquisitions are performed on the iris region of the target object to obtain multiple candidate iris texture images. Iris texture artifact filtering processing is performed on the multiple candidate iris texture images. Based on the local orientation field and anisotropic smoothing calculation of the iris texture structure, abnormal pixel regions corresponding to lens reflection and eyelash obstruction are removed to generate the iris texture image data of the target object.

4. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The process involves calling a pre-built dynamic fence topology learning model, using the fence distance change sequence and the iris texture image data as joint inputs for spatial consistency analysis, and generating identity fence association features for the target object, including: The fence distance change sequence is input into the spatial topology encoding branch of the dynamic fence topology learning model to extract the topological adjacency matrix between fence nodes and the abstract representation of the temporal motion trajectory of the target object on the fence node transfer path chain, and generate the fence spatial topology embedding vector. The iris texture image data is input into the iris texture encoding branch of the dynamic fence topology learning model. The iris texture is normalized and expanded and texture primitives are identified by multi-layer iris feature extraction operators to obtain the iris texture embedding vector. The fence spatial topology embedding vector and the iris texture embedding vector are input into the feature cross-fusion layer of the dynamic fence topology learning model to calculate the bidirectional interaction weights between the elements of each dimension in the fence spatial topology embedding vector and the elements of each dimension in the iris texture embedding vector. Based on the bidirectional interaction weights, a weighted outer product operation is performed on the fence spatial topology embedding vector and the iris texture embedding vector to generate a high-dimensional cross-fusion tensor representation, and the high-dimensional cross-fusion tensor representation is reconstructed into a fusion feature map aligned with the spatial channel dimension and the iris channel dimension. The fused feature map is input into the multi-scale decomposition convolutional layer of the dynamic fence topology learning model. The joint response patterns of fence spatial topology features and iris texture features under different receptive fields are extracted from the fused feature map to generate a multi-scale joint response feature set. The multi-scale joint response feature set is subjected to joint recalibration processing of channel attention and spatial attention to enhance the cross-channel collaborative expression of fence spatial topology features and iris texture features, and thus obtain collaborative enhancement features. The collaborative enhancement features are input into the identity mapping subnetwork of the dynamic fence topology learning model. The collaborative enhancement features are projected onto the identity feature space through a fully connected mapping to generate a preliminary identity mapping representation. The timestamp sequence of the target object passing through each electronic fence node is extracted from the fence node transfer path chain. The feature vectors corresponding to the timestamp sequence in the preliminary identity mapping representation are associated with the fence spatial location at the corresponding time point in the fence node transfer path chain to establish an association mapping table between identity feature values ​​and fence spatial locations. The continuous trajectory of the identity feature value of the target object as it moves within the fenced network is calculated based on the association mapping table. The spatial invariant component of this continuous trajectory is extracted as a spatially consistent identity representation. The spatially consistent identity representation and the spatial encoding of the fence node transfer path chain are then spliced ​​and fused to generate the identity fence association feature of the target object.

5. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The step of generating access control instructions for the target object based on the identity fence association features includes: Obtain the gate location coordinates and gate sensing coverage area of ​​the target access gate in the distributed fence network, and determine the spatial overlap area between the gate sensing coverage area and the sensing detection range of the adjacent electronic fence nodes. The spatially consistent identity representation in the identity fence association features is parsed, and the pre-stored authorized identity feature library is called to perform feature similarity comparison on the spatially consistent identity representation to generate an identity matching result identifier; When the identity matching result indicates a successful match, the fence node transfer path chain in the identity fence association feature is extracted, the predicted passage path segment of the target object from the current electronic fence node to the target access gate is calculated, and the expected arrival time interval of the target object to the target access gate is calculated based on the order of the electronic fence nodes passed through by the predicted passage path segment and the physical distance between the nodes. Based on the spatial overlap region and the expected arrival time interval, the access authorization time window of the target access gate is determined, and a gate pre-activation instruction containing the identity matching result identifier, the access authorization time window and the gate's orientation coordinates is generated. The gate pre-activation instruction is sent to the target access gate, so that the target access gate switches its gate status to a passable state within the access authorization time window. After the target access gate is switched to the passable state, real-time access sensing data at the target access gate is collected, the access trigger signal of the real-time access sensing data is monitored, the instantaneous access timestamp corresponding to the access trigger signal is extracted, and the time boundary is compared with the access authorization time window to obtain time window matching information; Based on the time window matching information and the passage completion signal returned by the passage safety sensor of the target passage gate, a passage event closure record is generated. The passage event closure record is then bound to the identity fence association feature of the target object to generate a passage control instruction for instructing the target passage gate to perform passage authorization operation.

6. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The method further includes: The node status of the electronic fence nodes in the distributed fence network is polled to obtain the acquisition window status parameters of the iris acquisition module and the sensing detection status parameters of the fence sensing module of each electronic fence node. Based on the acquisition window status parameters and the sensing detection status parameters, a single node status evaluation vector is constructed, and the single node status evaluation vectors of all electronic fence nodes are aggregated into a network status evaluation matrix. The network state evaluation matrix is ​​subjected to spatial filtering to detect target electronic fence nodes with abnormal state fluctuations. The abnormal state fluctuation features of the target electronic fence nodes are extracted and matched with the fault feature templates in the preset node fault mode library to determine the fault prediction type of the target electronic fence node. When the fault prediction type indicates that the iris acquisition module has failed, the topological position of the target electronic fence node in the fence node transfer path chain is extracted to determine the selection range of alternative nodes. Within the selection range of alternative nodes, the proximity to the target electronic fence node is calculated based on the node orientation coordinates of each candidate electronic fence node. The node health status is calculated based on the acquisition window status parameters, and the acquisition condition optimization level is calculated based on the iris acquisition suitability assessment value. The proximity, node health status, and acquisition condition optimization level are weighted and combined to generate an alternative node suitability score for each candidate electronic fence node, and the alternative node suitability scores are used for ranking. The alternative electronic fence nodes are determined according to the adaptability ranking of the alternative nodes, and the iris collection task is migrated from the target electronic fence node to the alternative electronic fence node. The logical topology diagram of the electronic fence nodes of the distributed fence network is updated, and the node identifier of the alternative electronic fence node is replaced with the node identifier of the target electronic fence node. The fence distance change data involving the target electronic fence node in the fence distance change sequence is recalculated. The fence distance change data is reconstructed based on the node orientation coordinates of the replacement electronic fence node. The fence distance change sequence of the target object is updated based on the reconstructed fence distance change data. The dynamic fence topology learning model is called again to generate the updated identity fence association features.

7. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The method further includes: When multiple target objects are detected to enter the same local area of ​​the distributed fence network at the same time, the fence distance change sequence and iris texture image data of each target object are acquired. Trajectory cross-analysis is performed on the fence distance change sequences of different target objects to extract the spatial distance change curves between the fence distance change sequences of different target objects on the common time axis. Local extremum detection processing is performed on the spatial distance change curves to extract the spatiotemporal points where the spatial distance between different target objects reaches a local minimum value, which are used as the intersection points of the object trajectories. A multi-object spatial interaction topology graph is constructed based on the intersection points of the object trajectories. The multi-object spatial interaction topology graph uses the target objects as nodes and the intersection points of the object trajectories as edges. The strength of the edges is determined by the local minimum value of the spatial distance. Community structure detection is performed on the multi-target spatial interaction topology map to identify group object combinations that form a clustered movement pattern within the fence network, and the group fence distance change sequence of the group object combinations is extracted. The iris texture image data of each target object in the group object combination is grouped and bound with the group fence distance change sequence to generate a group identity fence association feature set. A group access control instruction is generated based on the group identity fence association feature set. The group access control instruction is used to synchronously execute the access authorization operation of the target access gate for each target object in the group object combination.

8. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The method further includes: The sensing and detection historical data of the fence sensing modules of each electronic fence node in the distributed fence network are obtained within a preset historical period. A sensing and detection historical dataset containing the global sensing status of the fence network is constructed. Based on the sensing and detection historical dataset, the sensing trigger frequency distribution and sensing intensity fluctuation pattern of each electronic fence node in different time periods are extracted to generate fence sensing mode features. Based on the fence sensing mode characteristics of each electronic fence node, the electronic fence nodes in the distributed fence network are divided into functional areas, node clusters with similar fence sensing mode characteristics are identified, and the node clusters are mapped to the physical spatial layout of the distributed fence network to generate a functional area division map of the fence network. Based on the functional area division diagram of the fence network, the sensing frequency and sensing range parameters of the fence sensing module of the electronic fence node in different functional areas are adjusted accordingly. Based on the differentiated adjustment of the sensing frequency and sensing range parameters, the fence status perception strategy of the electronic fence nodes in each functional area is reconfigured to obtain an adaptive fence perception parameter configuration scheme, which is then sent to each electronic fence node through the node communication module to replace the current fence sensing module operating parameters.

9. The security management method based on iris recognition and electronic fence integration according to claim 1, characterized in that, The method further includes: The light source intensity adjustment record of the near-infrared illumination source and the focal length adjustment record of the optical focusing mechanism are obtained during the iris texture image data acquisition process. Based on the light source intensity adjustment record and the focal length adjustment record, the discernibility change curve of iris texture details in the iris texture image data is extracted. When the slope of the discernibility change curve is lower than the preset texture degradation threshold, the iris texture frequency domain response features of the iris texture image data of the corresponding time period are extracted, and the iris texture frequency domain response features are input into the pre-constructed iris texture quality degradation analysis model to generate iris texture quality degradation types. The iris texture quality degradation types include illumination uneven degradation type and focus drift degradation type. To address the aforementioned uneven illumination degradation type, a partitioned intensity compensation scheme for near-infrared illumination sources is generated, wherein the partitioned intensity compensation scheme sets differentiated near-infrared illumination intensities for different sub-regions of the iris region; For the aforementioned focus drift degradation type, a focus drift compensation trajectory for the optical focusing mechanism is generated. The focus drift compensation trajectory applies a compensation control amount opposite to the drift direction to the focusing drive signal of the optical focusing mechanism. The partition intensity compensation scheme and the focus drift compensation trajectory are fed back to the iris acquisition module of the corresponding electronic fence node to update the acquisition parameter configuration of the iris acquisition module.

10. A security management system based on iris recognition and electronic fence integration, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the security management method based on iris recognition and electronic fence integration as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images