An intelligent access control system

Abstract

This invention provides an intelligent access control system, relating to the field of intelligent security access control technology, comprising: an acquisition module for acquiring real-time access request data streams deployed at the entrance of the passage using radar probes, near-infrared multispectral cameras, and anti-tailgating buffer 3D structured light scanners; extracting spatial pose coordinate distribution parameters by parsing the access request data stream, and constructing a virtual access space envelope based on the spatial pose coordinate distribution parameters; and a mapping module for performing topological mesh discretization on the virtual access space envelope to obtain a micro-mesh array; and mapping the spatial mapping trajectory of the original multimodal biometric signals to the internal space of the micro-mesh array based on the node topological connection relationship of the micro-mesh array. This invention achieves accurate spatiotemporal alignment and spatial deformation compensation for multimodal access data.

Description

Technical Field

[0001] This invention relates to the field of intelligent security access control technology, and in particular to an intelligent access control system. Background Technology

[0002] In peak-hour, densely populated traffic scenarios, traditional multimodal intelligent access control systems employ a simple fusion architecture of facial recognition and infrared beam scanning for tailgating prevention. When two individuals pass through at extremely close range and there is strong backlight interference at the entrance, the accuracy of facial feature extraction by the visible light facial recognition module drops significantly due to the backlight effect. Simultaneously, the infrared beam scanning sensor can only detect the presence of heat sources within the passage and cannot distinguish between two independent human silhouettes. Because the system fails to achieve high-precision spatiotemporal alignment of multi-sensor data, there is a significant spatial misalignment and time delay between the movement trajectory of the person collected by the millimeter-wave radar and the facial image collected by the camera. This causes the system to mistakenly identify the fused data of two individuals as the passage data of a single authorized person and allow passage. Ultimately, an unauthorized person successfully tailgates into the office area, and the system is unable to detect the low-precision 3D-printed mask worn by the unauthorized person, creating a security risk of internal information leakage. This exposes the core technical defects of existing technologies, such as weak environmental anti-interference capability in complex environments, low multi-sensor fusion accuracy, poor tailgating detection accuracy, and susceptibility to forgery attacks. Summary of the Invention

[0003] This invention provides an intelligent access control system that achieves accurate spatiotemporal alignment and spatial deformation compensation for multimodal access data.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0005] Firstly, an intelligent access control system includes:

[0006] The acquisition module is used to acquire real-time access request data streams deployed on the access side of the channel, including radar probes, near-infrared multispectral cameras, and anti-tailgating buffer 3D structured light scanners; it extracts spatial pose coordinate distribution parameters by parsing the access request data streams, and constructs a virtual access space envelope based on the spatial pose coordinate distribution parameters;

[0007] The mapping module is used to perform topological mesh discretization on the virtual passage space envelope to obtain a micro-mesh array; based on the node topological connection relationship of the micro-mesh array, the spatial mapping trajectory of the original multimodal biometric signals is mapped to the internal space of the micro-mesh array;

[0008] The calculation module is used to convert the internal space of the micro-element mesh array into a geometric deformation field acting on the micro-element mesh array; to obtain the mesh strain energy by accumulating the stress distribution state of the geometric deformation field; to determine the target boundary nodes of the micro-element mesh array based on the spatial gradient distribution characteristics of the mesh strain energy; and to apply a virtual unit displacement vector field to the target boundary nodes to obtain the geometric deformation field and the virtual unit displacement vector field.

[0009] The correction module is used to analyze the local spatial offset of each micro-element grid by performing an inner product integral operation between the geometric deformation field and the virtual unit displacement vector field to obtain the feature projection offset correction value; the passage request data stream is corrected using the feature projection offset correction value to obtain the corrected passage feature vector set;

[0010] The determination module is used to comprehensively determine the access confidence of the corrected access feature vector set to obtain the determination result; based on the determination result, the physical drive control sequence and network access routing strategy of the access control actuator are mapped.

[0011] Furthermore, the system acquires real-time access request data streams from radar probes, near-infrared multispectral cameras, and 3D structured light scanners deployed at the channel entrance; it extracts spatial pose coordinate distribution parameters by parsing the access request data streams, and constructs a virtual access space envelope based on these parameters, including:

[0012] The spatial point cloud time series output by the radar probe at the entrance of the acquisition channel, the multi-band reflectivity map captured by the near-infrared multispectral camera, and the depth phase coding data obtained by the anti-tailgating buffer 3D structured light scanner are used to perform spatiotemporal reference alignment and communication protocol parsing on the spatial point cloud time series, multi-band reflectivity map and depth phase coding data, and then fused to obtain the passage request data stream.

[0013] Receive the passage request data stream, perform feature decoupling and three-dimensional coordinate mapping operations on the passage request data stream, remove environmental background interference components, and extract spatial pose coordinate distribution parameters including the spatial coordinates of the torso joint nodes of the passage personnel, the direction of the motion vector and the projection position of the center of gravity;

[0014] Receive spatial pose coordinate distribution parameters, use the set of extreme points of the outer contour of the spatial pose coordinate distribution parameters as the spatial constraint boundary, perform spatial surface interpolation, and obtain a closed and continuous virtual passage space envelope in which the volume dynamically occupied by the passage personnel is obtained.

[0015] Furthermore, a micro-mesh array is obtained by performing topological mesh discretization on the virtual access space envelope; based on the node topological connection relationship of the micro-mesh array, the spatial mapping trajectory of the original multimodal biometric signals is mapped to the internal space of the micro-mesh array, including:

[0016] Receive the virtual passage space envelope, and perform tetrahedral element subdivision and node numbering on the surface and internal volume of the virtual passage space envelope according to the preset mesh density threshold and curvature adaptive partitioning rules, to obtain a micro-element mesh array containing spatial coordinate index and topological adjacency matrix.

[0017] The system receives a micro-grid array, extracts the three-dimensional coordinate trajectory point set of the original multimodal biofeature signal within a continuous acquisition cycle, determines the spatial inclusion relationship between the three-dimensional coordinate trajectory point set and the spatial coordinate index of the micro-grid array, and maps the trajectory point sequence falling into the internal space of the micro-grid array as the spatial mapping trajectory of the original multimodal biofeature signal attached to the internal node of the corresponding grid unit based on the node topology connection relationship.

[0018] Furthermore, the internal space of the micro-mesh array is transformed into a geometric deformation field acting on the micro-mesh array; the mesh strain energy is obtained by accumulating the stress distribution state of the geometric deformation field, including:

[0019] The system receives a micro-mesh array and maps the passage-occupied volume represented by the spatial mapping trajectory of the original multimodal biometric signals to the internal space of the micro-mesh array. It then converts the temporal deformation increment of each mesh cell in the internal space into the node local deformation gradient tensor and constructs a geometric deformation field representing the evolution law of the relative position of the nodes based on the local deformation gradient tensor.

[0020] The nodal displacement bias and normal stress components of each grid element in the geometric deformation field are extracted. Based on the coupling mapping relationship between the nodal displacement bias and normal stress components, spatial domain integration is performed. The elastic potential energy components of each grid element are accumulated to obtain the grid strain energy, which is the degree of cumulative stress deformation of the entire space when people pass through the passage.

[0021] Furthermore, the target boundary nodes of the micro-element mesh array are determined based on the spatial gradient distribution characteristics of the mesh strain energy; a virtual unit displacement vector field is applied to the target boundary nodes to obtain the geometric deformation field and the virtual unit displacement vector field, including:

[0022] The grid strain energy is received, and gradient vector calculation is performed on the grid strain energy in the spatial topological domain of the micro-element grid array. The rate of change of strain energy between adjacent grid elements is analyzed to obtain the spatial gradient distribution characteristics of the grid strain energy.

[0023] Grid nodes whose gradient magnitudes are in the extreme range in the spatial gradient distribution characteristics are selected and identified as target boundary nodes. Using the target boundary nodes as excitation source points, virtual displacement excitation of a unit magnitude is injected along the topological connectivity path of the micro-element grid array to obtain a virtual unit displacement vector field that runs through the micro-element grid array. The virtual unit displacement vector field and the geometric deformation field are then spatially registered and aligned to obtain the geometric deformation field and the virtual unit displacement vector field.

[0024] Furthermore, the local spatial offset of each infinitesimal mesh is analyzed by the inner product integral operation of the geometric deformation field and the virtual unit displacement vector field to obtain the characteristic projection offset correction value, including:

[0025] Receive the geometric deformation field and the virtual unit displacement vector field, map the virtual unit displacement vector field into each grid cell of the micro-element grid array, and perform point-by-point vector matching with the geometric deformation field to obtain the matched vector field;

[0026] The inner product integration operation is performed on the matched vector field in the three-dimensional spatial domain of the micro-element mesh array, and the displacement component integral value of each mesh element in the preset projection axis is accumulated.

[0027] The local spatial offset of each micro-element mesh relative to the standard datum plane is analyzed based on the integral value of the displacement component.

[0028] Spatial topology weighted aggregation and normalization are performed on each local spatial offset to obtain the characteristic projection offset correction value of the spatial distortion degree of multi-source sensor data.

[0029] Furthermore, the passage request data stream is corrected using feature projection offset correction values ​​to obtain a corrected passage feature vector set, including:

[0030] Receive feature projection offset correction value and passage request data stream, and parse the feature projection offset correction value into spatial coordinate compensation parameters and time phase alignment parameters for each data channel;

[0031] Based on the spatial coordinate compensation parameters and the time phase alignment parameters, three-dimensional spatial coordinate remapping and time axis phase synchronization compensation are performed on the multimodal original signal in the passage request data stream to eliminate data misalignment caused by sensor array pose deviation and obtain the compensated multimodal original signal.

[0032] The compensated multimodal original signal is subjected to feature space dimensionality reduction and orthogonalization filtering to remove environmental coupling noise components and obtain the filtered multimodal feature components.

[0033] The filtered multimodal feature components are concatenated into tensors according to preset temporal rules and then encapsulated to obtain the corrected pass feature vector set.

[0034] Furthermore, the modified access feature vector set is combined with the access confidence determination result to obtain the determination result; based on the determination result, the physical drive control sequence and network access routing strategy of the access control actuator are mapped, including:

[0035] The system receives the corrected access feature vector set, extracts the biometric matching degree component, spatial trajectory compliance component, and environmental disturbance resistance component, inputs each component into the preset multidimensional threshold verification matrix to perform weighted fusion calculation, obtains the comprehensive access confidence index, and obtains the determination result of access permit level and risk intervention mark based on the comparison relationship between the comprehensive access confidence index and the preset authorization threshold.

[0036] Receive the judgment result, analyze the access permission level and risk intervention identifier, match the servo motor torque threshold, electromagnetic lock release delay parameter and anti-tailgating baffle closing acceleration command in the access control actuator based on the access permission level, and assemble the physical drive control sequence according to the time axis sequence;

[0037] Based on risk intervention identifiers and access permission levels, the virtual LAN partitioning strategy, data flow firewall filtering rules, and dynamic micro-segmentation access routing instructions for access terminals are dynamically obtained and encapsulated to obtain network access routing policies.

[0038] In a second aspect, a computing device includes:

[0039] One or more processors;

[0040] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to execute the system.

[0041] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, performs the system.

[0042] The above-described solution of the present invention has at least the following beneficial effects:

[0043] By employing a multimodal sensing fusion architecture that combines radar probes, near-infrared multispectral cameras, and anti-tailgating buffer 3D structured light scanners, along with virtual passage space envelope construction, topological mesh discretization, feature projection offset correction based on geometric deformation field and virtual work principle, and comprehensive passage confidence determination technology, this approach overcomes the core technical problems of existing technologies, such as weak environmental anti-interference capability, low spatiotemporal reference alignment accuracy of multi-sensor data, poor anti-tailgating detection accuracy, and susceptibility to spoofing attacks. This achieves accurate spatiotemporal alignment and spatial deformation compensation of multimodal passage data, improves identity recognition accuracy and anti-tailgating detection capability in complex environments, effectively resists various spoofing attacks, and ensures passage safety and enterprise information security. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of an intelligent access control system provided by an embodiment of the present invention.

[0045] Figure 2 This is a flowchart illustrating how an intelligent access control system, provided by an embodiment of the present invention, performs a comprehensive access confidence determination on a modified access feature vector set to obtain a determination result; and based on the determination result, maps the physical drive control sequence of the access control actuator and the network access routing strategy.

[0046] Figure 3 This is a bar chart illustrating the standardized performance comparison of multimodal characteristic indicators of an intelligent access control system provided by an embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram illustrating the process of constructing the virtual access space envelope and subdividing the micro-mesh in an intelligent access control system according to an embodiment of the present invention.

[0048] Figure 5 This is a schematic diagram of the control logic for access confidence classification and risk intervention in an intelligent access control system provided by an embodiment of the present invention. Detailed Implementation

[0049] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art.

[0050] like Figure 1 As shown, an embodiment of the present invention proposes an intelligent access control system, comprising:

[0051] The acquisition module is used to acquire real-time access request data streams deployed on the access side of the channel, including radar probes, near-infrared multispectral cameras, and anti-tailgating buffer 3D structured light scanners; it extracts spatial pose coordinate distribution parameters by parsing the access request data streams, and constructs a virtual access space envelope based on the spatial pose coordinate distribution parameters;

[0052] The mapping module is used to perform topological mesh discretization on the virtual passage space envelope to obtain a micro-mesh array; based on the node topological connection relationship of the micro-mesh array, the spatial mapping trajectory of the original multimodal biometric signals is mapped to the internal space of the micro-mesh array;

[0053] The calculation module is used to convert the internal space of the micro-element mesh array into a geometric deformation field acting on the micro-element mesh array; to obtain the mesh strain energy by accumulating the stress distribution state of the geometric deformation field; to determine the target boundary nodes of the micro-element mesh array based on the spatial gradient distribution characteristics of the mesh strain energy; and to apply a virtual unit displacement vector field to the target boundary nodes to obtain the geometric deformation field and the virtual unit displacement vector field.

[0054] The correction module is used to analyze the local spatial offset of each micro-element grid by performing an inner product integral operation between the geometric deformation field and the virtual unit displacement vector field to obtain the feature projection offset correction value; the passage request data stream is corrected using the feature projection offset correction value to obtain the corrected passage feature vector set;

[0055] The determination module is used to comprehensively determine the access confidence of the corrected access feature vector set to obtain the determination result; based on the determination result, the physical drive control sequence and network access routing strategy of the access control actuator are mapped.

[0056] In this embodiment of the invention, a multimodal sensing fusion architecture is adopted, which combines a radar probe, a near-infrared multispectral camera, and a three-dimensional structured light scanner for anti-tailgating buffers. This architecture is combined with the construction of a virtual passage space envelope, topological mesh discretization, multimodal biometric spatial mapping, geometric deformation field construction and mesh strain energy calculation, and the virtual unit displacement vector field based on the target boundary nodes. Based on the feature projection offset correction based on the virtual work principle and the comprehensive passage confidence judgment technology, this invention overcomes the core technical problems of existing technologies, such as weak environmental anti-interference ability, low spatiotemporal reference alignment accuracy of multi-sensor data, feature misalignment caused by spatial distortion of multi-source data, poor anti-tailgating detection accuracy, and susceptibility to spoofing attacks. This achieves accurate spatiotemporal alignment and spatial deformation compensation of multimodal passage data, improves the accuracy of identity recognition in complex environments and the detection capability of tailgating personnel at extremely close range, effectively resists various spoofing attacks, and realizes integrated linkage management of physical access control and network access permissions.

[0057] In a preferred embodiment of the present invention, a real-time data stream of passage request is acquired from a radar probe, a near-infrared multispectral camera, and a three-dimensional structured light scanner deployed on the channel entrance side; spatial pose coordinate distribution parameters are extracted by parsing the passage request data stream, and a virtual passage space envelope is constructed based on the spatial pose coordinate distribution parameters, including:

[0058] The system acquires a spatial point cloud time-series sequence output by the radar probe at the entrance of the acquisition channel, a multi-band reflectivity map captured by the near-infrared multispectral camera, and depth phase-coded data obtained by the 3D structured light scanner in the anti-tailgating buffer zone. It then performs spatiotemporal reference alignment and communication protocol parsing on the spatial point cloud time-series sequence, multi-band reflectivity map, and depth phase-coded data, fusing them to obtain a passage request data stream. Specifically, this involves continuously waking up the radar probe deployed at the top of the channel entrance, the near-infrared multispectral camera, and two 3D structured light scanners deployed on the left and right sides of the anti-tailgating buffer zone, ensuring all sensors are in continuous real-time acquisition mode. The radar probe continuously emits millimeter-wave signals and receives target reflection echoes. It calculates the target's spatial distance using the echo signal's time of flight and the target's velocity using Doppler frequency shift, outputting a spatial point cloud time-series sequence containing the target's 3D coordinates and velocity at fixed time intervals. The near-infrared multispectral camera simultaneously acquires reflected light signals within the channel in three preset near-infrared bands, converting the light intensity information of different bands into digital signals and outputting a multi-band reflectivity map containing multi-band pixel values. The 3D structured light scanner projects a sinusoidally encoded structured light pattern onto the anti-tailgating buffer, acquires the deformed structured light pattern reflected by the target, and outputs depth phase encoded data containing 3D depth information by grading the depth value of each pixel.

[0059] When personnel enter the sensor coverage area at the entrance of the passage, the system simultaneously receives raw data from three sensor devices and performs communication protocol parsing. It identifies the communication protocol type and data frame format of each sensor device, converts raw data in different formats such as TCP / IP, RS485, and CAN into a unified standardized data format within the system, and adds a device identifier and original timestamp field to each data frame.

[0060] After completing the communication protocol parsing, the system performs a spatiotemporal reference alignment operation for multimodal data. The system obtains the current timestamp of the system global clock, calculates the time offset of each sensor relative to the system global clock, and corrects the original timestamp of each data frame based on the calculated time offset, so that the timestamps of all data frames are unified under the system global clock reference. Linear interpolation is performed on sensor data with different sampling frequencies to ensure that the data frames of all sensor devices have the same time sampling interval, ensuring that the corresponding data frames from the three sensor devices can be acquired at any given time.

[0061] After completing spatiotemporal reference alignment and communication protocol parsing, the spatial point cloud time series, multi-band reflectivity map, and depth phase encoded data are fused to obtain a unified access request data stream. The access request data stream contains all access-related data from three different sensing devices at the same time and under the same time reference, and each data frame contains a unified global timestamp and device identifier.

[0062] The system receives a passage request data stream, performs feature decoupling and 3D coordinate mapping operations on it, removes environmental background interference components, and extracts spatial pose coordinate distribution parameters including the spatial coordinates of the torso joint nodes, motion vector direction, and center of gravity projection position of the person passing through. Specifically, this includes: receiving the passage request data stream and performing feature decoupling operations. Based on the data source and data type, the passage request data stream is split into three independent data channels: a radar point cloud data channel, a multispectral image data channel, and a structured light depth data channel. Each channel retains its corresponding global timestamp and spatial coordinate information.

[0063] A three-dimensional coordinate mapping operation is performed on the data from each data channel, converting the coordinates of each data point in the sensor's local coordinate system to coordinates in the system's global three-dimensional spatial coordinate system. After completing the three-dimensional coordinate mapping, an environmental background interference component removal operation is performed. Background data for multiple consecutive cycles when there are no people passing through the channel is collected in advance. When people pass through, the current passage request data stream is compared point by point with the static background data model. The absolute value of the difference between each data point and the corresponding background data point is calculated. When the absolute value of the difference is greater than a preset background difference threshold, the data point is determined to be a foreground data point, i.e., a data point belonging to the passing person. When the absolute value of the difference is less than or equal to the preset background difference threshold, the data point is determined to be a background data point, i.e., a data point belonging to the environmental background. All data points determined to be background data points are removed from the passage request data stream, completing the removal of the environmental background interference component.

[0064] From the passage request data stream stripped of environmental background interference components, spatial pose coordinate distribution parameters are extracted. Radar point cloud data and structured light depth data are fused to extract the complete three-dimensional torso contour information of the person passing through, and the global three-dimensional spatial coordinates of each key torso joint node are obtained.

[0065] Based on the joint node coordinates at two consecutive moments, the motion vector direction of each key trunk joint node is calculated. The joint node motion vector is the difference between the global three-dimensional coordinates of the joint node at the current moment and the global three-dimensional coordinates of the joint node at the previous moment. The weighted average of the global three-dimensional coordinates of all key trunk joint nodes of the person passing through is calculated to obtain the coordinates of the person's center of gravity.

[0066] The calculated center of gravity coordinates are vertically projected onto the horizontal plane where the passageway floor is located to obtain the projected position of the center of gravity. The system integrates the extracted spatial coordinates of all key trunk joint nodes, the motion vector directions of all joint nodes, and the projected position of the center of gravity to obtain complete spatial pose coordinate distribution parameters.

[0067] The system receives spatial pose coordinate distribution parameters, uses the set of extreme points of the outer contour of the spatial pose coordinate distribution parameters as the spatial constraint boundary, performs spatial surface interpolation, and obtains a closed, continuous virtual passage space envelope containing the volume dynamically occupied by the passage personnel. Specifically, this includes: receiving the obtained spatial pose coordinate distribution parameters, extracting the set of extreme points of the outer contour, traversing the coordinates of all key torso joint nodes in the spatial pose coordinate distribution parameters, and the coordinates of all passage personnel contour points extracted from the fused data, finding the maximum and minimum values ​​of all coordinate points in the x-axis, y-axis, and z-axis directions respectively, and using the set of the above six extreme points as the set of extreme points of the outer contour. This set of points determines the minimum circumscribed cuboid boundary of the volume dynamically occupied by the passage personnel.

[0068] Using the outer contour extreme point set as the spatial constraint boundary, a cubic B-spline surface fitting and reconstruction algorithm is called for spatial surface interpolation. The outer contour extreme point set is used as the initial control point set to construct a uniformly distributed cubic B-spline surface control mesh. The coefficient matrix of the B-spline surface is obtained by solving a system of linear equations, and then the coordinates of any point on the surface are calculated. The calculation formula is as follows: ;

[0069] in, Let be the three-dimensional coordinate vector of any point on the cubic B-spline surface. , They are respectively , cubic B-spline basis functions in the parameter direction, The coordinates of the control vertices formed by the extreme points of the outer contour. , They are respectively , The direction controls the number of vertices. For surface parameter coordinates, and For index variables, The cubic B-spline basis functions are: ;

[0070] in, For the node vector 1 node For quadratic B-spline basis functions, For the first A quadratic basis function, cubic B-spline basis functions and Similarly, the cubic B-spline basis functions can be obtained as follows: ;

[0071] in, for Direction Quadratic B-spline basis functions for Direction A quadratic B-spline basis function is used to calculate the surface coordinates of all sampling points within the spatial range defined by the set of extreme points of the outer contour, according to a preset sampling interval, to obtain a continuous and smooth spatial surface. The obtained spatial surface is then closed, and the edges of the surface are connected sequentially to form a closed and continuous three-dimensional surface. This three-dimensional surface completely surrounds the dynamic volume occupied by the pedestrian, and its shape and position are updated in real time with the movement of the pedestrian, which is the virtual passage space envelope.

[0072] In this embodiment of the invention, by employing a multi-source raw data synchronous acquisition technology using radar probes, near-infrared multispectral cameras, and 3D structured light scanners, combined with spatiotemporal reference alignment of multimodal data, unified parsing of communication protocols, feature decoupling and background interference removal, and surface fitting and reconstruction technology based on outer contour extreme point constraints, the technical problems of incompatibility of multi-sensor data formats, insufficient fusion accuracy due to spatiotemporal reference misalignment, severe background interference leading to distortion in human spatial feature extraction, and lack of a unified 3D spatial carrier to carry multimodal data in the prior art are overcome. Thus, standardized fusion and high-precision spatiotemporal synchronization of multi-source heterogeneous passage data are achieved, the core spatial motion and pose features of the passage personnel are accurately extracted, and a closed continuous virtual passage space envelope that fully represents the dynamic volume occupied by the passage personnel is constructed.

[0073] In a preferred embodiment of the present invention, a micro-mesh array is obtained by performing topological mesh discretization on the virtual access space envelope; the spatial mapping trajectory of the original multimodal biometric signals is mapped to the internal space of the micro-mesh array based on the node topological connection relationship of the micro-mesh array, including:

[0074] The system receives a virtual passageway envelope and, based on a preset mesh density threshold and curvature adaptive partitioning rule, performs tetrahedral element subdivision and node numbering on the surface and internal volume of the virtual passageway envelope, resulting in a micro-element mesh array containing spatial coordinate indices and a topological adjacency matrix. Specifically, this includes: receiving a closed, continuous virtual passageway envelope and initializing the core parameters for topological mesh partitioning. The system loads a preset mesh density threshold, which defines the maximum allowable side length of a single micro-element mesh cell, ensuring that the mesh accuracy meets the requirements of subsequent feature mapping and deformation calculations. Simultaneously, it loads a curvature adaptive partitioning rule, which stipulates that when the average surface curvature of a mesh cell exceeds a preset curvature threshold, the cell needs to be recursively subdivided to preserve the geometric details of complex surfaces.

[0075] Triangular meshing is performed on the surface of the virtual passage space envelope. All surface control points of the virtual passage space envelope are extracted, and an initial triangular surface mesh is obtained based on the control points. Each initial triangular cell is traversed, and its average surface curvature is calculated. The formula for calculating the average surface curvature is: the average surface curvature of a triangular cell is one-third of the sum of the curvature values ​​of the three vertices of the triangle. The calculated average surface curvature is compared with a preset curvature threshold. If the average surface curvature is greater than the preset curvature threshold, the triangular cell is bisected along its longest side to obtain two new triangular cells. The above curvature calculation and subdivision operation is repeated for all newly obtained triangular cells until the average surface curvature of all triangular cells is less than or equal to the preset curvature threshold, thus completing the surface meshing of the virtual passage space envelope.

[0076] After completing the surface meshing, the process expands into the internal volume of the virtual passageway envelope by performing tetrahedral element meshing. Using the surface triangular mesh as the boundary, the Delaunay tetrahedral meshing algorithm is employed to obtain the initial internal tetrahedral mesh. Each initial tetrahedral element is traversed, and the length of its longest side is calculated. If the longest side length is greater than a preset mesh density threshold, the tetrahedral element is subdivided into four new tetrahedral elements. This side length calculation and subdivision operation is repeated for all newly obtained tetrahedral elements until the longest side length of all tetrahedral elements is less than or equal to the preset mesh density threshold.

[0077] Perform a unified numbering and labeling operation on all obtained mesh nodes. Assign a unique integer number to each mesh node in the order of surface nodes first and then internal nodes to ensure that each node has a unique identifier in the entire micro-element mesh array. At the same time, assign a unique element number to each tetrahedral element and record the numbers of the four nodes contained in each tetrahedral element.

[0078] The spatial coordinate index of the micro-element mesh array is obtained, and a one-to-one correspondence between the mesh node number and the global 3D spatial coordinate of the node is established to form a spatial coordinate index table. Through the spatial coordinate index table, the global 3D spatial coordinates corresponding to any node can be quickly looked up based on the node number, and the nearest mesh node can be quickly located based on any global 3D spatial coordinate. The topological adjacency matrix of the micro-element mesh array is obtained. The topological adjacency matrix is ​​a two-dimensional matrix, and the rows and columns of the matrix correspond to the mesh node numbers. All elements of the topological adjacency matrix are initialized to 0. Then, each tetrahedral element is traversed, and the four node numbers of the element are extracted. For any two nodes belonging to the same tetrahedral element, the element at the intersection of the corresponding row and column in the topological adjacency matrix is ​​set to 1, indicating that there is a topological connection between the two nodes. After traversing all tetrahedral elements, the complete topological adjacency matrix is ​​obtained. This matrix completely describes the topological connection relationship between all nodes in the micro-element mesh array. The spatial coordinate index and the topological adjacency matrix are integrated to obtain the complete micro-element mesh array.

[0079] The process involves receiving a micro-grid array, extracting the 3D coordinate trajectory point set of the original multimodal biometric signals within continuous acquisition cycles, determining the spatial inclusion relationship between the 3D coordinate trajectory point set and the spatial coordinate index of the micro-grid array, and mapping the trajectory point sequence falling into the internal space of the micro-grid array as the spatial mapping trajectory of the original multimodal biometric signals attached to the internal nodes of the corresponding grid cells based on the node topology connection relationship. Specifically, this includes: receiving the micro-grid array; extracting the original multimodal biometric signals from the previously obtained access request data stream; the original multimodal biometric signals including radar echo feature signals, near-infrared multispectral facial feature signals, and 3D structured light morphology feature signals; extracting the global 3D spatial coordinates of each biometric point within multiple consecutive acquisition cycles to form the 3D coordinate trajectory of each biometric point; and integrating the 3D coordinate trajectories of all biometric points to obtain the 3D coordinate trajectory point set of the original multimodal biometric signals.

[0080] For each trajectory point in the 3D coordinate trajectory point set, a spatial inclusion relationship determination is performed to determine whether the trajectory point falls inside a tetrahedral element in the micro-element mesh array. For any trajectory point to be determined, based on its global 3D spatial coordinates, the nearest mesh node to the trajectory point is found through the spatial coordinate index table, and then all tetrahedral elements with topological connections to that node are traversed. For each candidate tetrahedral element, the volumes of the four smaller tetrahedra formed by the trajectory point and the four vertices of the tetrahedral element are calculated, and then the sum of the volumes of these four smaller tetrahedra is calculated. The sum of the volumes of the four smaller tetrahedra is compared with the volume of the original tetrahedral element. If the difference between the two is less than a preset volume error threshold, the trajectory point is determined to fall inside the tetrahedral element; if the difference is greater than or equal to the preset volume error threshold, the trajectory point is determined not to fall inside the tetrahedral element. The above process is repeated until a tetrahedral element containing the trajectory point is found, or all candidate tetrahedra are traversed.

[0081] For all trajectory points determined to fall within the space of the micro-element mesh array, a feature mapping operation is performed based on the node topology connection relationship. The four vertex numbers of the tetrahedral element containing the trajectory point are extracted, the spatial distance from the trajectory point to the four vertices is calculated, and the feature assignment weight of each vertex is calculated based on the distance from the trajectory point to the four vertices. The formula for calculating the feature assignment weight is as follows: ;

[0082] in, For the first Assign weights to the features of each vertex. =1, 2, 3 and 4, For the feature trajectory point to the th The spatial distance between the four grid vertices is the sum of the reciprocals of the corresponding distances of the four vertices.

[0083] The original multimodal biometric signal value carried by the trajectory point is assigned to the four vertices of the tetrahedral element according to the feature allocation weights calculated above. The final feature value of each vertex is the sum of the feature values ​​of all trajectory points mapped to that vertex. The spatial inclusion relationship determination and feature mapping operations are performed sequentially on all trajectory points in the three-dimensional coordinate trajectory point set. All trajectory point sequences falling into the internal space of the micro-element mesh array are mapped to the original multimodal biometric signal values ​​attached to the nodes inside the corresponding mesh element. The feature values ​​of each node at different times are arranged in chronological order to obtain the spatial mapping trajectory of the original multimodal biometric signal attached to the nodes of the micro-element mesh array.

[0084] In this embodiment of the invention, a virtual passage space envelope tetrahedral element partitioning technique based on a preset mesh density threshold and curvature adaptive partitioning rules, combined with node numbering and calibration, spatial coordinate indexing and topological adjacency matrix construction, and a multimodal biofeature three-dimensional coordinate trajectory point set mapping technique based on spatial inclusion relationship determination and node topological connection relationship, overcomes the technical problems in the prior art of lacking a unified discretization computing carrier for multimodal data, making it difficult to accurately represent the complex curved surfaces and internal spatial features of the human body, making it difficult for different modal biofeatures to accurately correspond in the same three-dimensional space, and losing topological relationships during feature mapping. Thus, high-precision discretization modeling of the virtual passage space is achieved, fully preserving the topological connection relationship and geometric details of the human body space, and mapping the original signals of multimodal biofeatures from independent sensor coordinates to the three-dimensional computing space of the micro-element mesh array, realizing accurate spatial location correspondence and topological association of different modal biofeatures.

[0085] In a preferred embodiment of the present invention, the internal space of the micro-element mesh array is converted into a geometric deformation field acting on the micro-element mesh array; the mesh strain energy is obtained by accumulating the stress distribution state of the geometric deformation field, including:

[0086] The process involves receiving a micro-grid array, mapping the passage volume represented by the spatial mapping trajectory of the original multimodal biosignal signals to the internal space of the micro-grid array, converting the temporal deformation increment of each grid cell in the internal space into a node local deformation gradient tensor, and constructing a geometric deformation field representing the evolution law of the relative position of the nodes based on the local deformation gradient tensor. Specifically, this includes receiving the micro-grid array obtained in the previous step, as well as the spatial mapping trajectory of the original multimodal biosignal signals attached to the nodes of the micro-grid array, and mapping the passage volume represented by the spatial mapping trajectory of the original multimodal biosignal signals to the internal space of the micro-grid array.

[0087] Traverse all grid nodes in the micro-element grid array and calculate the occupancy rate of each node at the current moment. The occupancy rate quantifies the degree to which the space where the node is located is occupied by pedestrians. The occupancy rate of a single node at the current moment is the quotient of the cumulative value of the multimodal biometric signal of the node at the current moment and the cumulative value of the multimodal biometric signal of the node at the historical maximum. Compare the occupancy rates of all nodes at the current moment with those of the previous moment and calculate the temporal displacement vector of each node. The temporal displacement vector is used to characterize the positional change of the node's space due to the movement of pedestrians. The calculation formula is as follows: ;

[0088] in, This represents the temporal displacement vector of a single node at the current time. The current node occupancy rate. The node's occupancy rate at the previous time step. The average side length of a single tetrahedral mesh cell. This is the direction vector of node motion.

[0089] The process iterates through all tetrahedral mesh elements in the micro-element mesh array, calculating the nodal local deformation gradient tensor for each element based on the temporal displacement vectors of its four vertices. The nodal local deformation gradient tensor describes the degree and direction of local deformation within the element's internal space. Its calculation process is as follows: For any tetrahedral mesh element, extract the x, y, and z displacement components of its four vertices, calculate the partial derivatives of each displacement component with respect to the x, y, and z axes, and arrange the nine partial derivatives in the order of the three-dimensional spatial coordinates to obtain the nodal local deformation gradient tensor of that tetrahedral mesh element. Then, arrange and combine the nodal local deformation gradient tensors of all tetrahedral mesh elements according to the spatial topology of the micro-element mesh array to construct a three-dimensional vector field covering the entire virtual passageway.

[0090] The nodal displacement bias and normal stress components of each mesh element in the geometric deformation field are extracted. Based on the coupling mapping relationship between the nodal displacement bias and normal stress components, spatial domain integration is performed, and the elastic potential energy components of each mesh element are accumulated to obtain the mesh strain energy representing the cumulative degree of spatial stress deformation when a person passes through the passage. Specifically, this includes: extracting the nodal displacement bias of each tetrahedral mesh element in the geometric deformation field. The nodal displacement bias refers to the portion of the actual displacement of the node that produces elastic deformation, and the rigid body displacement components that do not produce deformation need to be discarded. The calculation process is as follows: for any tetrahedral mesh element, the average translation vector of its four vertices is calculated as the overall translational displacement of the element, and the average rotation angle of its four vertices around the centroid of the element is calculated as the overall rotational displacement of the element; the actual temporal displacement vector of each node in the element is subtracted sequentially from the nodal displacement components corresponding to the overall translational displacement and rotational displacement of the element to obtain the displacement bias of that node.

[0091] Based on the nodal local deformation gradient tensor of each tetrahedral mesh element, the normal strain component of the element is calculated. The normal strain component is used to describe the degree of tensile or compressive deformation of the element in the three directions of x-axis, y-axis and z-axis. The normal strain component in a single direction is the deformation gradient component in the corresponding direction plus half of the transpose of the deformation gradient component in the corresponding direction.

[0092] Based on the preset material elastic parameters, the normal stress component of each tetrahedral mesh element is calculated. The normal stress component describes the magnitude of the internal forces acting on the element in three directions. The normal stress component in a single direction is the product of the preset elastic modulus and the normal strain component in the corresponding direction.

[0093] Based on the coupled mapping relationship between nodal displacement bias and normal stress components, spatial domain integration is performed to calculate the elastic potential energy component of a single tetrahedral mesh element. The elastic potential energy component is used to quantify the energy stored in a single element due to deformation. The calculation formula is as follows: ;

[0094] in, , and for , , Directional normal stress components, This represents the elastic potential energy component of a single tetrahedral mesh element. The volume of a single tetrahedral mesh element. , and for , , Directional normal strain component.

[0095] By traversing all tetrahedral mesh elements in the micro-element mesh array and summing the elastic potential energy components of all elements, the cumulative energy value of the deformation of the entire virtual passage space when a person passes through the passage is obtained, which is the mesh strain energy.

[0096] In this embodiment of the invention, the passage volume is mapped to the internal space of a micro-element mesh array, and the temporal deformation increment of the mesh element is converted into the local deformation gradient tensor of the node to construct a geometric deformation field. Combined with the extraction of the node displacement bias and normal stress components in the geometric deformation field, and the calculation method of mesh strain energy by accumulating the elastic potential energy components of each mesh element through spatial domain integration, this overcomes the technical problems in the prior art that it is difficult to convert the dynamic spatial occupation behavior of passage personnel into quantifiable physical field parameters, difficult to characterize the human body spatial boundary and deformation characteristics through spatial mechanical properties, and lack of means to quantitatively evaluate the degree of spatial distortion of multimodal data. Thus, it realizes the physical field quantitative characterization of the dynamic spatial occupation behavior of passage personnel, transforms the abstract multimodal data spatial distribution into a calculable geometric deformation field and mesh strain energy parameters, and accurately quantifies the cumulative degree of overall spatial force deformation during passage.

[0097] In a preferred embodiment of the present invention, the target boundary nodes of the micro-element mesh array are determined based on the spatial gradient distribution characteristics of the mesh strain energy; a virtual unit displacement vector field is applied to the target boundary nodes to obtain the geometric deformation field and the virtual unit displacement vector field, including:

[0098] The process involves receiving the mesh strain energy, performing gradient vector calculations on the mesh strain energy within the spatial topological domain of the micro-element mesh array, analyzing the rate of strain energy change between adjacent mesh elements, and obtaining the spatial gradient distribution characteristics of the mesh strain energy. Specifically, this includes: receiving the mesh strain energy calculated in the previous step, allocating the strain energy value of each tetrahedral mesh element to its four vertex nodes to obtain the nodal strain energy value of each mesh node; the nodal strain energy value of a single mesh node is the quotient of the sum of the strain energy values ​​of all tetrahedral mesh elements containing that node and the number of all tetrahedral mesh elements containing that node; and performing gradient vector calculations on the nodal strain energy values ​​of all mesh nodes within the spatial topological domain of the micro-element mesh array. For any mesh node to be calculated, all adjacent nodes with direct topological connections to that node are found using a topological adjacency matrix. The coordinate differences between the node and each adjacent node in the x-axis, y-axis, and z-axis directions, and the corresponding nodal strain energy differences, are calculated respectively.

[0099] The strain energy gradient components in the three directions are vector synthesized to obtain the strain energy gradient vector of the grid node. All grid nodes in the micro-element grid array are traversed, and the strain energy gradient vector and gradient magnitude of each node are calculated in turn. The strain energy change rate between adjacent grid elements is analyzed. The strain energy change rate is proportional to the strain energy gradient magnitude. The larger the gradient magnitude, the faster the strain energy changes at that location. The strain energy gradient vectors and gradient magnitudes of all grid nodes are arranged according to the spatial topology of the micro-element grid array to obtain the complete spatial gradient distribution characteristics of the grid strain energy.

[0100] Grid nodes with gradient amplitudes in the extreme range of the spatial gradient distribution characteristics are selected and identified as target boundary nodes. Using the target boundary nodes as excitation source points, virtual displacement excitation of a unit magnitude is injected along the topological connectivity path of the micro-element grid array to obtain a virtual unit displacement vector field that runs through the micro-element grid array. The virtual unit displacement vector field is then spatially registered and aligned with the geometric deformation field to obtain the geometric deformation field and the virtual unit displacement vector field. Specifically, this includes: sorting the strain energy gradient amplitudes of all grid nodes in descending order, determining the extreme range of the gradient amplitude, which is defined as the amplitude range corresponding to the top 5% of all nodes in terms of strain energy gradient amplitude, selecting all grid nodes whose strain energy gradient amplitudes are within this extreme range, and uniformly identifying them as target boundary nodes. Target boundary nodes are usually located at the interface between the passerby and the environment, as well as at the spatial interface between different passersby, accurately representing the mechanical boundary of different spatial regions.

[0101] Using each target boundary node as the excitation source point, a virtual displacement excitation of unit magnitude is injected along the topological connectivity path of the micro-element mesh array. The amplitude of the virtual displacement excitation is a preset unit displacement value, and the direction is along the outward normal of the target boundary node. The outward normal is calculated by cross product operation of the coordinates of the three adjacent nodes on the surface where the node is located.

[0102] For each target boundary node, a virtual displacement excitation is assigned to the node itself. All nodes directly connected to this node are found using the topological adjacency matrix. The virtual displacement response of the node is then calculated. The formula for calculating the virtual displacement response of a single adjacent node is as follows: ;

[0103] in, This represents the virtual displacement response of adjacent nodes. The virtual displacement value of the excitation source node, This represents the spatial distance between the source node and its neighboring node.

[0104] Repeat the above process, propagating the virtual displacement excitation layer by layer outward along the topological connectivity path until all grid nodes in the micro-element grid array have been traversed. For a grid node simultaneously affected by multiple excitation source points, its final virtual displacement value is the vector sum of the virtual displacement responses generated by all excitation source points at that node. Arranging the virtual displacement values ​​of all grid nodes according to the spatial topological position of the micro-element grid array yields a three-dimensional vector field that runs through the entire micro-element grid array, which is the virtual unit displacement vector field.

[0105] The obtained virtual unit displacement vector field is spatially registered and aligned with the previously constructed geometric deformation field. The grid node coordinates of the two fields are matched one-to-one to ensure that the two fields have the same spatial coordinate reference. The timestamps of the two fields are aligned. After the registration and alignment are completed, a geometric deformation field and a virtual unit displacement vector field with completely unified spatial and temporal references are obtained.

[0106] In this embodiment of the invention, by performing gradient vector calculation on the strain energy of the mesh within the spatial topological domain of the micro-element mesh array, and analyzing the rate of change of strain energy to obtain the spatial gradient distribution characteristics, combined with the method of selecting nodes in the extreme range of gradient amplitude as target boundary nodes; injecting virtual displacement excitation along the topological connected path with the target boundary nodes as excitation source points to obtain a virtual unit displacement vector field, and aligning the virtual unit displacement vector field with the geometric deformation field in a spatial field, this invention overcomes the technical problems in the prior art, such as the difficulty in accurately identifying human body boundaries through spatial mechanical properties; the difficulty in effectively distinguishing multiple human body boundaries that are extremely close together; the susceptibility of traditional boundary recognition to occlusion and environmental interference; and the lack of a standardized reference field to support subsequent feature offset correction. This invention achieves high-precision human body boundary recognition based on mechanical properties, accurately locates the spatial interface between different passing personnel, obtains a unified and calculable virtual unit displacement vector field, and completes accurate registration with the geometric deformation field.

[0107] In a preferred embodiment of the present invention, the local spatial offset of each micro-element grid is analyzed by the inner product integral operation of the geometric deformation field and the virtual unit displacement vector field to obtain the characteristic projection offset correction value, including:

[0108] The process involves receiving the geometric deformation field and the dummy unit displacement vector field, mapping the dummy unit displacement vector field onto each grid cell of the micro-element mesh array, and performing point-by-point vector matching with the geometric deformation field to obtain the matched vector field. Specifically, this includes receiving the geometric deformation field and the dummy unit displacement vector field obtained in the previous step, which have completely unified spatial and temporal references, and mapping the dummy unit displacement vector field onto the four vertex nodes of each tetrahedral grid cell according to the nodes of the micro-element mesh array. Since the two fields have already completed spatial field registration and alignment, all grid nodes have completely identical spatial coordinates and timestamps in both fields.

[0109] Traverse all grid nodes in the micro-element grid array, extract the displacement vector in the geometric deformation field and the virtual displacement vector in the virtual unit displacement vector field for each node, match the two vectors of the same node one-to-one to obtain the matching vector pair of the node, arrange the matching vector pairs of all nodes according to the spatial topology of the micro-element grid array to obtain the matched vector field.

[0110] The inner product integration operation is performed on the matched vector field within the three-dimensional spatial domain of the micro-element mesh array. The integral values ​​of the displacement components of each mesh element along the preset projection axis are accumulated. Specifically, this includes: performing the inner product integration operation on the matched vector field within the entire three-dimensional spatial domain of the micro-element mesh array; traversing all tetrahedral mesh elements; extracting the matching vector pairs of the four vertex nodes for each element; and calculating the inner product of the two vectors at each vertex node.

[0111] The average value of the vector inner product of the four vertex nodes of the tetrahedral mesh element is calculated as the vector inner product value at the center of the element. The inner product integral value of the element in the entire spatial domain is calculated using the element center integration method. The inner product integral value of a single tetrahedral mesh element is the product of the vector inner product value at the center of the element and the volume of the tetrahedral mesh element. The displacement component integral values ​​of each tetrahedral mesh element in the three preset projection axes (x-axis, y-axis, z-axis) are calculated separately. For each projection axis, the displacement component of that axis is extracted separately, and the above inner product calculation and integration process is repeated to obtain the displacement component integral value of the element in the corresponding axis. All tetrahedral mesh elements are traversed, and the displacement component integral value of each element in the three projection axes is calculated in turn to complete the inner product integration operation in the entire three-dimensional spatial domain.

[0112] The local spatial offset of each micro-element mesh relative to the standard travel reference plane is analyzed based on the integral value of the displacement component. Specifically, it includes: analyzing the local spatial offset of each micro-element mesh relative to the standard travel reference plane based on the integral value of the displacement component of each tetrahedral mesh element. The standard travel reference plane is pre-calibrated under ideal conditions without any sensing distortion and is used to characterize the spatial reference under normal travel conditions.

[0113] For any tetrahedral mesh element, calculate its average displacement value along the three projection axes; perform vector synthesis of the average displacement values ​​along the three axes to obtain the local spatial offset of the tetrahedral mesh element relative to the standard passage reference plane. The local spatial offset accurately quantifies the degree of distortion of the multi-source sensor data at the spatial location of the element, and the magnitude of the offset is proportional to the degree of data distortion; traverse all tetrahedral mesh elements and calculate the local spatial offset of each element in turn to obtain the local spatial offset distribution covering the entire virtual passage space.

[0114] Spatial topological weighted aggregation and normalization are performed on each local spatial offset to obtain the feature projection offset correction value for the spatial distortion degree of multi-source sensor data. Specifically, this includes: performing spatial topological weighted aggregation on the local spatial offsets of all micro-grids. The weight of the spatial topological weighted aggregation is related to the spatial distance from the grid cell to the nearest target boundary node. The closer the grid cell is to the target boundary node, the higher its data distortion degree, and the greater its corresponding weight. The formula for calculating the weight of a single grid cell is: ;

[0115] in, For the first The weight value of each grid cell. For the first The square of the spatial distance from each grid cell to the nearest target boundary node.

[0116] Calculate the sum of the products of the local spatial offsets of all grid cells and their corresponding weight values, then divide by the sum of all weight values ​​to obtain the overall spatial offset after weighted aggregation. The calculation formula is as follows: ;

[0117] in, For the first Local spatial offset of each grid cell This represents the total number of grid cells in the micro-element grid array. This represents the overall spatial offset after weighted aggregation.

[0118] The weighted aggregated overall spatial offset is normalized and mapped to a value range of 0 to 1. The minimum and maximum overall spatial offsets are obtained in advance through statistical analysis of a large amount of sample data, corresponding to the overall spatial offsets in the undistorted state and the maximum distortion state, respectively. The final feature projection offset correction value comprehensively quantifies the degree of overall spatial distortion of the multi-source sensor data.

[0119] In this embodiment of the invention, a technique is used to perform point-by-point vector matching between a virtual unit displacement vector field and a geometric deformation field, and to calculate the integral value of the displacement component of each grid cell by performing inner product integration in the three-dimensional spatial domain. This is combined with a method that analyzes the local spatial offset of each micro-grid based on the integral value of the displacement component, and performs spatial topological weighted aggregation and normalization on the local spatial offset. Therefore, this method overcomes the technical problems in the prior art, such as the difficulty in accurately quantifying the spatial distortion degree of multi-source sensor data based on physical principles, the difficulty in analyzing local spatial offset differences grid by grid, and the difficulty of adapting traditional global unified correction methods to the distribution of local spatial distortion. This method achieves high-precision feature projection offset quantification calculation based on the principle of virtual work, accurately analyzes the local distortion degree at different spatial locations, and obtains feature projection offset correction values ​​that are adaptive to local spatial distortion differences.

[0120] In a preferred embodiment of the present invention, the passage request data stream is corrected using a feature projection offset correction value to obtain a corrected passage feature vector set, including:

[0121] The system receives feature projection offset correction values ​​and access request data streams, and parses the feature projection offset correction values ​​into spatial coordinate compensation parameters and temporal phase alignment parameters for each data channel. Specifically, this includes: receiving the feature projection offset correction values ​​obtained in the previous step, as well as the initially acquired raw access request data stream; loading a correction value mapping table pre-established through multi-scene calibration experiments, which records the spatial distortion coefficients and temporal distortion coefficients of each sensor data channel corresponding to different feature projection offset correction values; substituting the feature projection offset correction values ​​into the correction value mapping table, and calculating the spatial distortion coefficients and temporal distortion coefficients of the radar point cloud data channel, near-infrared multispectral image data channel, and three-dimensional structured light depth data channel through linear interpolation.

[0122] Based on the spatial distortion coefficients of each channel, the spatial coordinate compensation parameters of the corresponding channel are calculated. The spatial coordinate compensation parameters include six components: x-axis translation, y-axis translation, z-axis translation, x-axis rotation angle, y-axis rotation angle, and z-axis rotation angle.

[0123] Based on the time distortion coefficients of each channel, the time phase alignment parameters of the corresponding channels are calculated. The time phase alignment parameter is the time delay compensation amount of the channel relative to the global clock. The time phase alignment parameter of a single channel is the product of the time distortion coefficient of the channel and the preset reference time delay amount. The calculated spatial coordinate compensation parameters of the three data channels are integrated with the time phase alignment parameters to obtain a complete set of multimodal data compensation parameters.

[0124] Based on the spatial coordinate compensation parameters and the time phase alignment parameters, three-dimensional spatial coordinate remapping and time axis phase synchronization compensation are performed on the multimodal original signals in the access request data stream to eliminate data misalignment caused by sensor array pose deviation and obtain the compensated multimodal original signals. Specifically, this includes extracting three independent data channels from the original access request data stream: radar point cloud original signal, near-infrared multispectral original signal, and three-dimensional structured light original signal.

[0125] A three-dimensional spatial coordinate remapping operation is performed on each data channel. For any original data point in the data channel, its original three-dimensional coordinates in the sensor's local coordinate system are extracted. A rotation transformation is performed on the original coordinates according to the rotation angle parameter of the channel, and a translation transformation is performed on the rotated coordinates according to the translation parameter of the channel. This process is repeated for all original data points in each data channel, and the above rotation and translation transformations are performed sequentially to complete the three-dimensional spatial coordinate remapping of all data points, thereby eliminating spatial data misalignment caused by sensor array installation pose deviations and environmental factors.

[0126] A time-axis phase synchronization compensation operation is performed on each data channel. For any given data channel, linear interpolation is performed on the original data sequence based on its time-phase alignment parameters. The formula for linear interpolation is: ;

[0127] in, The data value after interpolation at the target time. This is the original data value from the previous moment. This represents the original data value at the next moment. For the target time that needs interpolation, This refers to the time point corresponding to the previous moment. This is the time point corresponding to the next moment.

[0128] By using linear interpolation, the data sequences of all data channels are unified to the same time sampling point, ensuring that the three data channels have corresponding data values ​​at any given time, thus eliminating the time delay differences between different sensing devices. The data from the three data channels after spatial coordinate remapping and time axis phase synchronization compensation are integrated to obtain the compensated multimodal original signal.

[0129] The compensated multimodal original signal undergoes feature space dimensionality reduction and orthogonal filtering to remove environmental coupling noise components, yielding filtered multimodal feature components. Specifically, this includes: performing feature space dimensionality reduction by expanding the original data from the radar point cloud channel, near-infrared multispectral data channel, and 3D structured light depth data channel into one-dimensional feature vectors according to spatial point order; then concatenating the one-dimensional feature vectors from the three channels sequentially to form an initial high-dimensional original feature vector; and calculating the covariance matrix corresponding to the high-dimensional original feature vector. The covariance matrix characterizes the linear correlation between each feature dimension, where the element in the i-th row and j-th column is calculated as follows: ;

[0130] In the formula, The element in the i-th row and j-th column of the covariance matrix. This represents the total number of samples. Let i be the i-th feature value of the k-th sample. For the first The j-th feature value of a sample, The sample mean of the i-th feature value. It is the sample average value of the j-th feature.

[0131] After calculating the covariance matrix, solve for all eigenvalues ​​and their corresponding eigenvectors. Sort all eigenvalues ​​in descending order of value, select the eigenvectors corresponding to the top 8 largest eigenvalues, and arrange these eigenvectors column-wise to form the feature projection matrix. Perform matrix multiplication between the high-dimensional original eigenvectors and the feature projection matrix. The result is the low-dimensional eigenvector after removing redundant dimensions. The calculation formula is as follows: ;

[0132] In the formula, The reduced-dimensional feature vectors are the result of dimensionality reduction. This is the initial high-dimensional original feature vector. is the characteristic projection matrix.

[0133] Gram-Schmidt orthogonalization filtering is performed on the low-dimensional feature vectors to transform the low-dimensional feature vector group with dimensional correlation into a pairwise orthogonal feature vector group. At the same time, environmental noise-related components are separated. The orthogonalization processing rules are as follows: the first orthogonal feature vector is directly taken from the first original low-dimensional feature vector; the second orthogonal feature vector is the second original low-dimensional feature vector minus its projection component on the first orthogonal feature vector; the third orthogonal feature vector is the third original low-dimensional feature vector minus its projection components on the first and second orthogonal feature vectors in turn. The orthogonalization transformation of all low-dimensional feature vectors is completed according to this recursive rule. After the orthogonalization processing is completed, a preset noise threshold of 0.1 is applied to remove orthogonal feature components with feature values ​​less than 0.1. These components mainly correspond to environmental coupling noise such as strong light backlight and equipment electromagnetic interference. The remaining effective orthogonal feature components are integrated to finally obtain the filtered multimodal feature components.

[0134] The filtered multimodal feature components are concatenated into tensors according to a preset temporal rule, and then encapsulated to obtain a corrected common feature vector set. Specifically, this includes: loading a preset temporal rule, which specifies the concatenation order and weight allocation of multimodal feature components at different acquisition times; arranging the filtered multimodal feature components from a preset number of consecutive acquisition times in chronological order according to the preset temporal rule; and multiplying each feature component at each acquisition time by the corresponding temporal weight coefficient, which varies with the time distance from the current time, with a larger weight coefficient for closer proximity to the current time.

[0135] All weighted feature components are concatenated into a one-dimensional tensor to form a unified long vector. Normalization is then performed on the concatenated long vector, mapping all feature values ​​to a range of 0 to 1. The normalized long vector is then encapsulated to obtain the final corrected common feature vector set.

[0136] In this embodiment of the invention, the technique of resolving the feature projection offset correction value into spatial coordinate compensation parameters and time phase alignment parameters for each data channel, combined with the three-dimensional spatial coordinate remapping and time axis phase synchronization compensation of the multimodal original signal, feature space dimensionality reduction and orthogonal filtering, and the method of splicing and encapsulating multimodal feature component tensors according to preset time sequence rules, overcomes the technical problems in the prior art of spatial misalignment and temporal asynchrony of multi-source sensor data, difficulty in effectively eliminating sensor array installation pose deviation, severe environmental coupling noise interference, and difficulty in standardizing and unifying the encapsulation of multimodal features. Thus, it achieves accurate spatial and temporal compensation of multimodal original data, completely eliminates data distortion caused by sensor array pose deviation and environmental factors, effectively removes environmental coupling noise components, and obtains a standardized and high-quality corrected pass feature vector set.

[0137] like Figure 2 As shown, in another preferred embodiment of the present invention, the modified access feature vector set is comprehensively analyzed to determine the access confidence level, thereby obtaining a determination result; based on the determination result, the physical drive control sequence and network access routing strategy of the access control actuator are mapped, including:

[0138] The system receives the corrected access feature vector set, extracts the biometric matching degree component, spatial trajectory compliance component, and environmental disturbance resistance component, and inputs each component into a preset multidimensional threshold verification matrix for weighted fusion calculation to obtain a comprehensive access confidence index. Based on the comparison between the comprehensive access confidence index and the preset authorization threshold, the system determines the access permit level and risk intervention indicator. Specifically, this includes: extracting the biometric matching degree component; comparing the near-infrared multispectral facial feature data and three-dimensional structured light morphological feature data in the access feature vector set with the corresponding standard biometric templates stored in the authorized personnel database dimension by dimension; calculating the near-infrared multispectral facial feature similarity and the three-dimensional structured light morphological feature similarity; and then obtaining the biometric matching degree component through weighted calculation. The calculation formula is as follows: ;

[0139] In the formula, For biometric matching degree components, For near-infrared multispectral facial feature similarity, The similarity is between 0 and 1 for the three-dimensional structured light morphology features. The higher the value, the better the feature matching.

[0140] The spatial trajectory compliance component is extracted by drawing the center-of-gravity projection positions of pedestrians from multiple consecutive data collection times in the corrected traffic feature vector set. These positions are then combined to form the actual traffic trajectory. This actual trajectory is compared point-by-point with a pre-calibrated standard single-person traffic trajectory. The average distance deviation between the two trajectories is statistically calculated, and the spatial trajectory compliance component is then obtained through calculation. The calculation formula is as follows: ;

[0141] In the formula, For spatial trajectory compliance components, The average distance deviation between the actual trajectory and the standard trajectory. To determine the maximum permissible trajectory deviation, if the calculated result is less than 0, then 0 is directly taken as the final value of the spatial trajectory compliance component.

[0142] The environmental immunity component is extracted, and combined with the overall deformation degree of the previously constructed geometric deformation field and the total amount of environmental noise stripped out during the filtering stage, the environmental immunity component is calculated using the following formula: ;

[0143] In the formula, For environmental immunity components, The degree of overall geometric deformation, Total environmental noise The maximum permissible total noise level is determined by the range of 0 to 1 for both the overall geometric deformation and the total environmental noise level.

[0144] Substituting the biometric matching component, spatial trajectory compliance component, and environmental immunity component obtained above into the preset weighting rule, the comprehensive access confidence index is calculated using the following formula: ;

[0145] In the formula, To achieve a comprehensive confidence index, As a biometric matching index, This is an indicator of spatial trajectory compliance. As an environmental immunity index, the comprehensive passability confidence index is compared with the preset authorization threshold to classify the passability level and match the corresponding risk intervention sign. The passability level and risk intervention sign are integrated to obtain the final judgment result.

[0146] The comprehensive pass confidence index is compared with preset three-level authorization thresholds, where the first authorization threshold is set to 0.9, the second authorization threshold is set to 0.75, and the third authorization threshold is set to 0.6. If the comprehensive pass confidence index is greater than or equal to 0.9, it is determined as a level 1 pass permit, corresponding to a no-risk intervention label; if the index is greater than or equal to 0.75 and less than 0.9, it is determined as a level 2 pass permit, corresponding to a low-risk intervention label; if the index is greater than or equal to 0.6 and less than 0.75, it is determined as a level 3 pass permit, corresponding to a medium-risk intervention label; if the index is less than 0.6, it is determined as a prohibition on passage, corresponding to a high-risk intervention label. The pass permit level and risk intervention label are integrated to form the final judgment result.

[0147] The system receives the judgment result, parses the access permission level and risk intervention identifier, and matches the servo motor torque threshold, electromagnetic lock release delay parameter, and anti-tailgating wing closing acceleration command in the access control actuator based on the access permission level. It then assembles the physical drive control sequence according to the timeline. Specifically, this includes: parsing the access permission level and risk intervention identifier in the judgment result; loading a pre-established actuator parameter mapping table, which records the access control actuator parameters corresponding to different access permission levels and risk intervention identifiers; and matching the corresponding servo motor torque threshold, electromagnetic lock release delay parameter, and anti-tailgating wing closing acceleration command from the mapping table based on the parsed access permission level and risk intervention identifier. The lower the access permission level and the higher the risk intervention identifier level, the larger the corresponding servo motor torque threshold, the shorter the electromagnetic lock release delay, and the greater the anti-tailgating wing closing acceleration, thus improving anti-tailgating capability and emergency response speed.

[0148] According to the preset timeline, the matched actuator parameters are assembled into a physical drive control sequence. The timing assembly rules for the physical drive control sequence are as follows: at time 0, an electromagnetic lock release command is sent, carrying the matched electromagnetic lock release delay parameter; 50 milliseconds after the electromagnetic lock release is completed, a servo motor drive command is sent, carrying the matched servo motor torque threshold, driving the anti-tailgating baffle to open; after the anti-tailgating baffle is fully open and maintains a preset passage time, an anti-tailgating baffle closing command is sent, carrying the matched anti-tailgating baffle closing acceleration command; after the anti-tailgating baffle is fully closed, an electromagnetic lock locking command is sent. The assembled physical drive control sequence is sent to the access control actuator to control the access control actuator to complete the corresponding actions.

[0149] Based on risk intervention identifiers and access permission levels, the system dynamically generates virtual LAN partitioning policies, data flow firewall filtering rules, and dynamic micro-segmentation access routing instructions for the accessing terminals. These are then encapsulated into network access routing policies, specifically including: dynamically generating network access control rules for the personnel using the access permit and risk intervention identifier; for personnel with Level 1 access permits and no risk intervention identifiers, obtaining full-authority internal network access rules; for personnel with Level 2 access permits and low-risk intervention identifiers, obtaining internal network rules restricting access to sensitive areas; for personnel with Level 3 access permits and medium-risk intervention identifiers, obtaining network rules allowing access only to public service areas; and for personnel prohibited from access and with high-risk intervention identifiers, obtaining rules completely denying network access.

[0150] Based on the obtained network access control rules, corresponding virtual LAN partitioning policies, data flow firewall filtering rules, and dynamic micro-segmentation access routing instructions are dynamically generated. The virtual LAN partitioning policy is used to assign the terminal devices of the users to the corresponding virtual LANs; the data flow firewall filtering rules are used to restrict the network access ports and protocols of the terminal devices; and the dynamic micro-segmentation access routing instructions are used to achieve physical isolation between the terminal devices and internal sensitive servers.

[0151] The obtained virtual LAN partitioning strategy, data flow firewall filtering rules, and dynamic micro-segmentation access routing instructions are integrated and encapsulated to obtain the final network access routing strategy.

[0152] In this embodiment of the invention, a hierarchical judgment technique is used to extract three dimensions: biometric matching degree, spatial trajectory compliance, and environmental immunity. These components are then weighted and fused using a multi-dimensional threshold verification matrix to calculate a comprehensive access confidence index. This is combined with an integrated linkage method that matches access control actuator parameters based on access permission levels and assembles physical drive control sequences in a time sequence. Finally, a network access control strategy is dynamically obtained based on risk intervention identifiers and access permission levels. This overcomes the technical problems in existing technologies, such as the susceptibility of single biometric judgment to interference leading to high false positive and false negative rates, the disconnect between physical access control and network access permissions, the difficulty in dynamically adjusting control strategies according to risk levels, and the difficulty in adapting fixed actuator parameters to different access scenarios. As a result, accurate access permission judgment with multi-dimensional fusion is achieved, while simultaneously realizing the integrated linkage between physical access control and network security management.

[0153] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0154] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0155] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0156] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0157] Example: Experimental Verification of Intelligent Access Control System

[0158] To further verify the technical effectiveness of the intelligent access control system proposed in this application, a 30-day field experiment was conducted at the main entrance of a large commercial complex. The experimental scenario was a standard pedestrian passage 1.2 meters wide and 2.5 meters long, where multimodal sensing and acquisition equipment was deployed, including millimeter-wave radar probes, near-infrared multispectral cameras, and 3D structured light scanners.

[0159] Step 2: Topological mesh discretization and multimodal biometric feature mapping

[0160] The virtual passage space is discretized into an 8×8 micro-grid array, with each grid cell measuring 0.15 m × 0.31 m. The system maps multimodal sensing data to the corresponding micro-grid, including characteristic parameters such as radar point cloud density, infrared spectral reflectivity, and structured light depth accuracy.

[0161] Experimental data show that the mean value of radar point cloud density features is 0.847, the mean value of infrared spectral reflectance features is 0.923, the mean value of structured light depth accuracy features is 0.956, the spatiotemporal alignment accuracy reaches 99.2%, and the feature extraction completeness is 91.8%. The distribution of each feature parameter is as follows: Figure 3 As shown.

[0162] Step 3: Geometric Deformation Field Construction and Mesh Strain Energy Calculation

[0163] Based on a micro-element mesh array, the system constructs a geometric deformation field and calculates the strain energy of each mesh element. During the passage of personnel, the strain energy of the core passage area mesh gradually accumulates from an initial value of 0.15, reaching a steady-state value of approximately 0.68 after about 2.5 seconds; the steady-state strain energy of the edge area mesh is approximately 0.38; and the steady-state strain energy of the buffer zone mesh is approximately 0.17.

[0164] The system sets the target boundary node screening threshold to 0.65. When the mesh strain energy exceeds this threshold, the corresponding mesh node is marked as a target boundary node for feature projection offset correction calculation. Figure 4 The evolution trend of strain energy over time in three types of regional grids is shown.

[0165] Step 4: Calculation of Feature Projection Offset Correction Value

[0166] A dummy unit displacement vector field is applied to the target boundary nodes, and the local spatial offset is analyzed through inner product integration. The system uses an iterative optimization algorithm to calculate the feature projection offset correction value, which converges after approximately 50 iterations.

[0167] Experimental results show that the final offset correction value is 0.38 cm in the X-axis direction, 0.42 cm in the Y-axis direction, and 0.35 cm in the Z-axis direction, with a combined offset correction value of 0.39 cm. The accuracy of the corrected passage request data stream is significantly improved, providing a reliable data foundation for subsequent passage confidence determination. Figure 5 The convergence process of the offset correction values ​​in each dimension is shown.

[0168] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent access control system, characterized by, include: The acquisition module is used to acquire real-time data streams of passage requests from radar probes, near-infrared multispectral cameras, and anti-tailgating buffer 3D structured light scanners deployed on the entrance side of the passage. Spatial pose coordinate distribution parameters are extracted by parsing the passage request data stream, and a virtual passage space envelope is constructed based on the spatial pose coordinate distribution parameters; The mapping module is used to perform topological mesh discretization on the virtual passage space envelope to obtain a micro-mesh array; Based on the node topology connection relationship of the micro-mesh array, the spatial mapping trajectory of the original multimodal biometric signals is mapped to the internal space of the micro-mesh array; The computation module is used to convert the internal space of the micro-mesh array into a geometric deformation field acting on the micro-mesh array; The mesh strain energy is obtained by accumulating the stress distribution state of the geometric deformation field; the target boundary nodes of the micro-element mesh array are determined based on the spatial gradient distribution characteristics of the mesh strain energy. A virtual unit displacement vector field is applied to the target boundary node to obtain the geometric deformation field and the virtual unit displacement vector field; The correction module is used to analyze the local spatial offset of each micro-element mesh through the inner product integration operation of the geometric deformation field and the virtual unit displacement vector field, so as to obtain the characteristic projection offset correction value. The passage request data stream is corrected by the feature projection offset correction value to obtain the corrected passage feature vector set; The determination module is used to comprehensively determine the passage confidence result of the corrected passage feature vector set to obtain the determination result. Based on the judgment results, the physical drive control sequence and network access routing strategy of the access control actuator are mapped.

2. The intelligent access control system of claim 1, wherein, Acquire real-time data streams of passage requests from radar probes, near-infrared multispectral cameras, and 3D structured light scanners deployed at the entrance of the passage; Spatial pose coordinate distribution parameters are extracted by parsing the passage request data stream. A virtual passage space envelope is constructed based on these parameters, including: The spatial point cloud time series output by the radar probe at the entrance of the acquisition channel, the multi-band reflectivity map captured by the near-infrared multispectral camera, and the depth phase coding data obtained by the anti-tailgating buffer 3D structured light scanner are used to perform spatiotemporal reference alignment and communication protocol parsing on the spatial point cloud time series, multi-band reflectivity map and depth phase coding data, and then fused to obtain the passage request data stream. Receive the passage request data stream, perform feature decoupling and three-dimensional coordinate mapping operations on the passage request data stream, remove environmental background interference components, and extract spatial pose coordinate distribution parameters including the spatial coordinates of the torso joint nodes of the passage personnel, the direction of the motion vector and the projection position of the center of gravity; Receive spatial pose coordinate distribution parameters, use the set of extreme points of the outer contour of the spatial pose coordinate distribution parameters as the spatial constraint boundary, perform spatial surface interpolation, and obtain a closed and continuous virtual passage space envelope in which the volume dynamically occupied by the passage personnel is obtained.

3. The intelligent access control system of claim 2, wherein, Perform topological mesh discretization on the virtual passage space envelope to obtain a micro-element mesh array; The node topology connections based on the micro-mesh array map the spatial mapping trajectory of the original multimodal biometric signals to the internal space of the micro-mesh array, including: Receive the virtual passage space envelope, and perform tetrahedral element subdivision and node numbering on the surface and internal volume of the virtual passage space envelope according to the preset mesh density threshold and curvature adaptive partitioning rules, to obtain a micro-element mesh array containing spatial coordinate index and topological adjacency matrix. The system receives a micro-grid array, extracts the three-dimensional coordinate trajectory point set of the original multimodal biofeature signal within a continuous acquisition cycle, determines the spatial inclusion relationship between the three-dimensional coordinate trajectory point set and the spatial coordinate index of the micro-grid array, and maps the trajectory point sequence falling into the internal space of the micro-grid array as the spatial mapping trajectory of the original multimodal biofeature signal attached to the internal node of the corresponding grid unit based on the node topology connection relationship.

4. The intelligent access control system of claim 3, wherein, The internal space of the micro-mesh array is converted into a geometric deformation field acting on the micro-mesh array; The mesh strain energy is obtained by accumulating the stress distribution state of the geometric deformation field, including: The system receives a micro-mesh array and maps the passage-occupied volume represented by the spatial mapping trajectory of the original multimodal biometric signals to the internal space of the micro-mesh array. It then converts the temporal deformation increment of each mesh cell in the internal space into the node local deformation gradient tensor and constructs a geometric deformation field representing the evolution law of the relative position of the nodes based on the local deformation gradient tensor. The nodal displacement bias and normal stress components of each grid element in the geometric deformation field are extracted. Based on the coupling mapping relationship between the nodal displacement bias and normal stress components, spatial domain integration is performed. The elastic potential energy components of each grid element are accumulated to obtain the grid strain energy, which is the degree of cumulative stress deformation of the entire space when people pass through the passage.

5. The intelligent access control system of claim 4, wherein, The target boundary nodes of the micro-element mesh array are determined based on the spatial gradient distribution characteristics of the mesh strain energy. Applying a virtual unit displacement vector field to the target boundary nodes yields the geometric deformation field and the virtual unit displacement vector field, including: The grid strain energy is received, and gradient vector calculation is performed on the grid strain energy in the spatial topological domain of the micro-element grid array. The rate of change of strain energy between adjacent grid elements is analyzed to obtain the spatial gradient distribution characteristics of the grid strain energy. Filter and identify grid nodes whose gradient magnitude is in the extreme value range in the spatial gradient distribution characteristics as target boundary nodes; Using the target boundary node as the excitation source point, a virtual displacement excitation of unit magnitude is injected along the topological connectivity path of the micro-element mesh array to obtain a virtual unit displacement vector field that runs through the micro-element mesh array. The virtual unit displacement vector field is then spatially registered and aligned with the geometric deformation field to obtain the geometric deformation field and the virtual unit displacement vector field.

6. The intelligent access control system of claim 5, wherein, The local spatial offset of each infinitesimal mesh element is analyzed by performing an inner product integral operation between the geometric deformation field and the virtual unit displacement vector field to obtain the characteristic projection offset correction value, including: Receive the geometric deformation field and the virtual unit displacement vector field, map the virtual unit displacement vector field into each grid cell of the micro-element grid array, and perform point-by-point vector matching with the geometric deformation field to obtain the matched vector field; The inner product integration operation is performed on the matched vector field in the three-dimensional spatial domain of the micro-element mesh array, and the displacement component integral value of each mesh element in the preset projection axis is accumulated. The local spatial offset of each micro-element mesh relative to the standard datum plane is analyzed based on the integral value of the displacement component. Spatial topology weighted aggregation and normalization are performed on each local spatial offset to obtain the characteristic projection offset correction value of the spatial distortion degree of multi-source sensor data.

7. The intelligent access system of claim 6, wherein, The passage request data stream is corrected using feature projection offset correction values ​​to obtain a corrected passage feature vector set, including: Receive feature projection offset correction value and passage request data stream, and parse the feature projection offset correction value into spatial coordinate compensation parameters and time phase alignment parameters for each data channel; Based on the spatial coordinate compensation parameters and the time phase alignment parameters, three-dimensional spatial coordinate remapping and time axis phase synchronization compensation are performed on the multimodal original signal in the passage request data stream to eliminate data misalignment caused by sensor array pose deviation and obtain the compensated multimodal original signal. The compensated multimodal original signal is subjected to feature space dimensionality reduction and orthogonalization filtering to remove environmental coupling noise components and obtain the filtered multimodal feature components. The filtered multimodal feature components are concatenated into tensors according to preset temporal rules and then encapsulated to obtain the corrected pass feature vector set.

8. The intelligent access control system of claim 7, wherein, The corrected access feature vector set is combined with the access confidence determination result to obtain the determination result; Based on the judgment result, the physical drive control sequence and network access routing strategy of the access control actuator are mapped, including: The system receives the corrected access feature vector set, extracts the biometric matching degree component, spatial trajectory compliance component, and environmental disturbance resistance component, inputs each component into the preset multidimensional threshold verification matrix to perform weighted fusion calculation, obtains the comprehensive access confidence index, and obtains the determination result of access permit level and risk intervention mark based on the comparison relationship between the comprehensive access confidence index and the preset authorization threshold. Receive the judgment result, analyze the access permission level and risk intervention identifier, match the servo motor torque threshold, electromagnetic lock release delay parameter and anti-tailgating baffle closing acceleration command in the access control actuator based on the access permission level, and assemble the physical drive control sequence according to the time axis sequence; Based on risk intervention identifiers and access permission levels, the virtual LAN partitioning strategy, data flow firewall filtering rules, and dynamic micro-segmentation access routing instructions for access terminals are dynamically obtained and encapsulated to obtain network access routing policies.

9. A computing device, comprising: include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, performs the system as described in any one of claims 1 to 8.

Images