Weak network environment access control method and system, storage medium and program product

Abstract

A method, system, storage medium, and program product for access control in weak network environments are disclosed, relating to the field of industrial information and data processing. The method includes: acquiring a target image; performing feature matching processing on the target image and a local feature library to obtain target candidate identifiers and corresponding initial matching scores; when the initial matching score falls within a preset suspicious interval, extracting the current fine-grained feature map with spatial topological structure from the target image; performing difference analysis to obtain local difference feature data; acquiring auxiliary recognition data through the local auxiliary acquisition module of the access control management system during a preset suspicious waiting period; performing cross-fusion verification on the local difference feature data, auxiliary recognition data, and auxiliary registration data to obtain a compensation judgment result; and outputting an access control opening signal to control access control opening when the compensation judgment result matches a preset release state. Implementing this application can improve access control efficiency in weak network environments.

Description

Technical Field

[0001] This application relates to the field of industrial information and data processing, and in particular to a method, system, storage medium and program product for access control in a weak network environment. Background Technology

[0002] With the expansion of smart parks and security management, access control systems are being deployed in complex physical environments such as remote factories, underground parking lots, and temporary construction sites. Cloud servers, with their powerful computing capabilities, can handle blurred or occluded features; while local comparison algorithms are typically lightweight and have poor recognition capabilities. However, network signals are easily affected by physical obstructions or insufficient base station coverage, leading to weak network conditions such as high latency, frequent packet loss, or intermittent disconnections, making data interaction between access control terminals and the cloud extremely unstable.

[0003] In related technologies, a cloud-edge collaborative solution is typically adopted to achieve high-quality and efficient access control functionality. This involves two aspects: firstly, encapsulating feature data and sending concurrent requests to a cloud server, using a high-performance cloud-based model to process blurred or occluded features and obtain high-precision comparison results; secondly, simultaneously calling a lightweight algorithm locally on the access control terminal to concurrently compare the collected feature data with a pre-downloaded local feature library. The return status from these two links determines the final access permission based on the first valid comparison result, thus reducing user waiting time.

[0004] However, under conditions of high latency and frequent packet loss, the return time of cloud comparison results is generally significantly slower than local processing, causing the system to rely almost entirely on the local lightweight comparison link when actually allowing passage. Without the support of cloud computing power to dynamically correct complex features, the system will generate a large number of false rejection results when faced with the degradation of the physical characteristics of on-site personnel, reducing passage efficiency. Summary of the Invention

[0005] This application provides a method, system, storage medium, and program product for access control in weak network environments, which can improve access control efficiency in weak network environments.

[0006] Firstly, this application provides a method for access control in a weak network environment, applied to an access control management system. The method includes: acquiring a target image; performing feature matching processing on the target image and a local feature library to obtain a target candidate identifier and its corresponding initial matching score; when the initial matching score falls within a preset suspicious interval, extracting a current fine-grained feature map with spatial topology from the target image; performing difference analysis on the current fine-grained feature map spatially based on the baseline feature map of the target candidate identifier, extracting feature components exceeding a significant difference threshold and their corresponding spatial location indices to obtain local difference feature data; acquiring auxiliary recognition data through a local auxiliary acquisition module of the access control management system during a preset suspicious waiting period; performing cross-fusion verification on the local difference feature data, auxiliary recognition data, and auxiliary registration data associated with the target candidate identifier to obtain a compensation determination result; and outputting an access control opening signal to control access control opening when the compensation determination result matches a preset release state.

[0007] In the above embodiments, the access control management system extracts the current feature data that falls into the suspicious interval and performs difference analysis with the benchmark feature map to obtain local difference feature data, so that the comparison process focuses on the feature change area; combined with the auxiliary recognition data obtained by the local auxiliary acquisition module, cross-fusion verification is performed to make up for the limitations of single image recognition in weak network and feature degradation, and improve the accuracy and passage efficiency of local comparison.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of outputting an access control opening signal to control access control opening when the compensation determination result matches the preset release state, the method further includes: responding to the access control opening signal, performing an association mapping between local difference feature data and target candidate identifiers in a local feature library to generate difference cache templates; obtaining lifecycle tags for multiple difference cache templates; the lifecycle tags include the last trigger timestamp and / or cumulative hit frequency; and removing the target difference cache template from the local feature library when the idle time of the target difference cache template exceeds a preset aging threshold or the cumulative hit frequency is lower than a preset active threshold, based on the lifecycle tags.

[0009] In the above embodiments, the access control management system generates a difference cache template from local difference feature data and performs association mapping, enabling the system to remember temporary feature changes of users; dynamic cleanup of the difference cache template based on lifecycle tags avoids invalid occupation of local storage space and ensures the retrieval efficiency of the local feature library.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of performing association mapping between local difference feature data and target candidate identifier in a local feature library in response to an access control opening signal to generate a difference cache template, the method further includes: obtaining multiple difference cache templates associated with the target candidate identifier in the local feature library; performing feature fusion on the multiple difference cache templates according to the cumulative hit frequency to obtain feature incremental update data; and updating the baseline feature map of the target candidate identifier in the local feature library based on the feature incremental update data.

[0011] In the above embodiments, the access control management system performs feature fusion on the high-frequency hit differential cache templates, extracts stable feature incremental update data, and updates the baseline feature map accordingly. This enables the local feature library to adaptively evolve with the long-term changes in the user's appearance, reducing the false rejection rate caused by feature aging.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after performing differential analysis on the current fine-grained feature map spatially based on the benchmark feature map of the target candidate identifier, extracting feature components exceeding the significance threshold of difference and their corresponding spatial location indices to obtain local differential feature data, the method further includes: obtaining the connectivity state parameters of the current network communication link and calculating the network quality assessment index; compressing and encoding the local differential feature data based on the network quality assessment index to generate a lightweight feature verification package; sending the lightweight feature verification package to the cloud server for comparison, and establishing a corresponding asynchronous listening callback task to obtain the comparison verification results returned by the cloud server.

[0013] In the above embodiments, the access control management system adaptively compresses and encodes local difference feature data based on network quality assessment indicators to generate a lightweight feature verification package; it interacts with the cloud through asynchronous listening callback tasks, maximizing the use of limited bandwidth in a weak network environment without blocking the local verification process, thus ensuring the availability of the high-computing-power verification link in the cloud.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after sending the lightweight feature verification package to the cloud server for comparison and establishing a corresponding asynchronous listening callback task to obtain the comparison verification result returned by the cloud server, the method further includes: when the comparison verification result is received through the asynchronous listening callback task, parsing the cloud feature repair data carried in the comparison verification result; and overwriting the benchmark feature map of the target candidate identifier in the local feature library based on the cloud feature repair data to obtain the local benchmark correction map.

[0015] In the above embodiments, the access control management system analyzes the feature repair data transmitted back from the cloud and overwrites the local benchmark feature map, thus bringing the high-precision feature processing capabilities of the cloud down to the local area. This enables the local feature library to obtain high-quality feature correction when the network is intermittently connected due to weak network conditions, thereby improving the reliability of subsequent pure local comparisons.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the step of compressing and encoding local differential feature data based on network quality assessment metrics to generate a lightweight feature verification package specifically includes: determining the target dimensionality reduction scale and the target quantization step size according to the network quality assessment metrics; performing dimensionality reduction processing on the local differential feature data according to the target dimensionality reduction scale based on a preset dimensionality reduction projection matrix to obtain a differential dimensionality reduction vector; discretizing and encoding the differential dimensionality reduction vector based on the target quantization step size, and attaching the index information of the target candidate identifier, and encapsulating it to generate a lightweight feature verification package.

[0017] In the above embodiments, the access control management system dynamically adjusts the dimensionality reduction scale and quantization step size according to the network quality, performs dimensionality reduction processing and discretization encoding on the differential features, and controls the data volume of the verification packet while retaining the core differential features, so that the data transmission strategy is highly matched with the current network carrying capacity.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the step of outputting an access control opening signal to control access control opening when the compensation determination result matches the preset release state specifically includes: when the compensation determination result matches the preset release state, extracting the spatial distribution features of local difference feature data in the target acquisition image and calculating the difference clustering degree; when the difference clustering degree is lower than a preset safety threshold, generating an anti-counterfeiting challenge instruction; driving the interaction module of the access control terminal to execute the anti-counterfeiting challenge instruction to collect liveness response features, and outputting an access control opening signal to control access control opening after the liveness response features pass verification; when the difference clustering degree is greater than or equal to the preset safety threshold, outputting an access control opening signal to control access control opening.

[0019] In the above embodiments, the access control management system calculates the degree of difference clustering by analyzing the spatial distribution characteristics of local difference feature data, thereby assessing potential counterfeiting risks; for high-risk scenarios with low clustering, a liveness detection challenge is introduced, which, while ensuring passage efficiency, strictly prevents illegal intrusion by splicing or forging local features.

[0020] In a second aspect, embodiments of this application provide an access control management system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the access control management system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an access control management system, cause the access control management system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on an access control management system, cause the access control management system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the access control management system provided in the second aspect, the computer storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. By employing the technical means of extracting current feature data and performing difference analysis with the benchmark feature map, and combining it with auxiliary recognition data for cross-fusion verification, the access control management system can complete the comprehensive judgment of multi-dimensional features locally without relying on cloud computing power when there is doubt in the initial comparison. This effectively solves the problem of high false rejection rate of local lightweight algorithms caused by cloud response lag in weak network environments in related technologies, and achieves high accuracy and high efficiency of access control in weak network environments.

[0026] 2. By adopting the technical means of generating difference cache templates from local difference feature data and dynamically managing and updating features based on lifecycle tags, the access control management system can continuously capture and adapt to the gradual changes in the appearance features of target personnel. This effectively solves the problem of continuous decline in recognition performance caused by feature aging in local feature libraries in related technologies, and realizes the adaptive evolution of local benchmark feature maps and long-term stable recognition performance.

[0027] 3. By adopting a technical means of dynamically compressing and encoding local difference feature data based on network quality assessment indicators and obtaining cloud feature repair data through asynchronous listening callback tasks, the access control management system can maintain non-blocking communication with the cloud under extremely low bandwidth. This effectively solves the problem of communication link paralysis caused by data packet congestion in weak network environments in related technologies, and realizes the reliable distribution and overwriting of high-precision feature repair capabilities from the cloud to the local feature library. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a method for access control in a weak network environment, as described in this application.

[0029] Figure 2 This is another flowchart illustrating the access control method for weak network environments in this application embodiment;

[0030] Figure 3 This is a schematic diagram of the physical device structure of an access control management system in an embodiment of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] In the weak network environment access control scenario involved in the embodiments of this application, the access control management system is usually deployed in areas with complex physical environments such as remote factory areas or underground parking lots.

[0034] In the implementation of related technologies, access control terminals rely on a cloud-edge collaborative architecture, that is, simultaneously initiating high-computing-power comparisons in the cloud and lightweight comparisons locally. However, in the aforementioned complex physical environment, network communication links are limited, resulting in high latency or frequent packet loss, causing severe delays in the return time of cloud comparison results or even request timeouts.

[0035] At this point, the access control terminal can only rely on a local lightweight algorithm to output the judgment result. Due to the limited number of parameters and computing power of the local model, when the person passing through has facial feature degradation, partial physical occlusion, or sudden changes in ambient lighting, the local lightweight algorithm cannot extract enough effective features for accurate matching, thus generating a large number of false rejection results.

[0036] Meanwhile, due to the persistently weak network conditions, the local feature library cannot obtain high-precision feature updates from the cloud in a timely manner, resulting in the continuous accumulation of feature aging on the local side. This causes the recognition pass rate of the access control system to decline progressively, severely restricting the passage efficiency.

[0037] The following describes the process of the method provided in this implementation. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a method for access control in a weak network environment, as described in this application.

[0038] S101. Acquire the target image, perform feature matching processing on the target image and the local feature library to obtain the target candidate identifier and the corresponding initial matching score.

[0039] The target image refers to the digital image containing the visual features of the person passing through, captured by the image acquisition module of the access control system during the current access event. The local feature library refers to the set of registered person feature templates stored locally on the access control terminal. The target candidate identifier represents the unique identification code of the registered person in the local feature library that has the highest similarity to the target image. The initial matching score represents the calculated similarity value between the global features of the target image and the template corresponding to the target candidate identifier.

[0040] Specifically, when the access control system's sensing module detects a person approaching, it triggers an image acquisition command to obtain the target image. The access control system then calls a local lightweight feature extraction network to convert the target image into a multi-dimensional feature vector. Using Euclidean distance or cosine similarity algorithms, this multi-dimensional feature vector is compared against all registered templates in the local feature library. The access control system selects the registration template with the highest similarity, extracts its corresponding identity code as a target candidate identifier, and records the calculated highest similarity value as the initial matching score in memory, providing basic data support for subsequent permission determination.

[0041] In some embodiments, feature matching between the target image and the local feature library can be achieved in various ways: Optionally, the access control system performs grayscale and normalization processing on the target image, extracts local binary pattern features of the image; calculates Hamming distance between the extracted local binary pattern features and template data in the local feature library; sorts the identifiers in the local feature library in ascending order according to the distance calculation results, and selects the identifier with the smallest distance as the target candidate identifier. Optionally, the access control system inputs the target image into a pre-trained lightweight convolutional neural network model and outputs a floating-point feature vector; calculates the dot product between the floating-point feature vector and the feature matrix in the local feature library through matrix multiplication; extracts the maximum value of the dot product result as the initial matching score, and maps the corresponding identifier as the target candidate identifier. It is understood that other hardware-accelerated matching methods based on edge computing architecture can also be used to achieve feature matching processing, which is not limited here.

[0042] S102. When the initial matching score falls into the preset doubtful interval, extract the current fine-grained feature map with spatial topological structure from the target acquisition image.

[0043] The preset doubtful range refers to a similarity score range set by the system, such as 60 to 80 points, indicating that the matching result is neither completely confirmed nor completely denied. Spatial topology represents the geometric arrangement and relative positional relationships between local features in an image, such as the relative distances between facial features or the connectivity of textures. The current fine-grained feature map is used to represent the high-resolution feature matrix containing rich local details and high-frequency information extracted from the target image.

[0044] This step is triggered when the access control system completes preliminary global feature matching and finds the matching score to be in an uncertain state. Specifically, the access control system first locates the valid face or access credential region in the target image and calls a pre-deployed fine-grained feature extraction algorithm or local attention network to perform multi-scale convolution operations on this region. During this process, the system strictly preserves the two-dimensional or three-dimensional spatial dimensional information of the feature map, ensuring that the extracted features not only contain local texture details but also retain the relative spatial coordinates of these details in the original image. This generates a current fine-grained feature map with spatial topology, providing high-resolution data support for subsequent accurate local difference comparison.

[0045] In some embodiments, the extraction of the current fine-grained feature map can be achieved in several ways: Optionally, the access control system inputs the target acquisition image into a multi-branch convolutional neural network; extracts local feature maps at different scales through a feature pyramid structure; concatenates the feature maps at each scale along the channel dimension while preserving the spatial dimension to generate the current fine-grained feature map. Optionally, the access control system uses an image segmentation algorithm to divide the target acquisition image into multiple semantic blocks; independently calculates a local binary pattern histogram for each semantic block; and combines the histograms of each block into a feature matrix according to the spatial topological order of the original image to generate the current fine-grained feature map. It is understood that other methods based on visual Transformer networks to extract local attention patches can also be used to extract the current fine-grained feature map, which is not limited here. It should be noted that during the extraction process, the system performs adaptive illumination compensation preprocessing on the image to ensure that the fine-grained features are not affected by sudden changes in ambient light.

[0046] In some embodiments, individuals may wear large masks, industrial safety helmets, or carry heavy goods, obscuring their faces and causing large-area physical occlusion in the target image, leading to the failure of conventional full-face feature extraction. The access control system can utilize a pre-trained local attention feature extraction subnetwork to identify and mask occluded areas (such as mask or helmet areas) using image semantic segmentation technology. This allows the receptive field of feature extraction to focus on the unoccluded, valid facial areas (such as around the eyes and forehead), thereby extracting a fine-grained feature map that removes occlusion interference while preserving the effective spatial topology. This ensures that even in situations where physical occlusion is suspected, the feature extraction process can still output usable high-dimensional spatial data.

[0047] S103. Based on the baseline feature map of the target candidate identifier, perform differential analysis on the current fine-grained feature map in each spatial dimension, extract the feature components that exceed the significance threshold of difference and their corresponding spatial location indices, and obtain local differential feature data.

[0048] The baseline feature map refers to a high-dimensional feature matrix pre-stored in the local feature library, representing the standard appearance of the target candidate identifier. The significance threshold represents the numerical limit used to determine whether the difference between two feature components is statistically or physically significant. A feature component refers to the specific value of the feature map at a particular channel or location. The spatial location index represents the coordinate position of the feature component in the two-dimensional or three-dimensional space of the feature map. Local difference feature data represents the set of all extracted feature components exceeding the threshold and their coordinates.

[0049] After the system successfully extracts the current fine-grained feature map, this step is executed to accurately locate local changes in a person's appearance. Specifically, the access control management system spatially aligns the current fine-grained feature map with the baseline feature map in memory. Then, the system traverses each spatial dimension and channel of the map, calculating the absolute difference or relative rate of change of the feature components at corresponding locations. The system compares the calculated differences one by one with a preset significance threshold, filtering out minor fluctuations caused by normal lighting or slight posture changes, and accurately selecting feature components whose differences exceed the threshold. Finally, the system records these significantly changed feature components and their corresponding spatial location indices, packaging them to generate local difference feature data, allowing subsequent verification to focus entirely on areas where substantial changes have occurred.

[0050] In some embodiments, local difference feature data extraction can be achieved in several ways: Optionally, the access control system performs element-wise matrix subtraction on the current fine-grained feature map and the baseline feature map; takes the absolute value of the residual matrix after subtraction; filters out elements with absolute values ​​less than the difference significance threshold using a mask matrix, and extracts the remaining elements and non-zero coordinates to generate local difference feature data. Optionally, the access control system calculates the cosine distance between the two feature maps at various spatial locations; converts the cosine distance into a difference probability distribution map; and uses a connected component analysis algorithm to extract connected region pixels with probabilities greater than the difference significance threshold and their coordinates to generate local difference feature data. It is understood that other methods based on feature channel attention mechanisms to calculate difference weights can also be used to extract local difference feature data, and this is not limited here.

[0051] In some embodiments, when the access control terminal is in a low-light environment such as at night or in a basement, the image sensor gain is amplified, which may introduce a large number of high-frequency noise points such as snowflakes into the target image. This can lead to the image noise being misidentified as facial texture changes in the difference analysis results. To address this, the access control management system can introduce a spatial connectivity verification mechanism when performing spatial dimension-wise difference analysis. Specifically, after extracting feature components that exceed the difference significance threshold, the system maps these feature components to a two-dimensional spatial location index and performs morphological opening operations or connected component analysis to remove isolated high-frequency difference points with an area smaller than a preset noise area threshold. Only effective feature components exhibiting a certain clustering pattern are retained, thereby obtaining pure local difference feature data and effectively avoiding interference from low-light noise in subsequent verification.

[0052] S104. During the preset waiting period for suspected cases, auxiliary identification data is obtained through the local auxiliary acquisition module of the access control management system.

[0053] The preset doubtful waiting period represents the maximum time window during which the access control system allows the current verification process to continue while waiting for additional data input after determining that the initial matching score is questionable. The local auxiliary acquisition module refers to non-visual sensors or secondary visual sensors integrated into the access control terminal. Auxiliary identification data represents signals or data output by the local auxiliary acquisition module used to provide multimodal identity verification.

[0054] Specifically, upon triggering the discrepancy analysis, the access control system starts an internal timer to calculate a preset suspicious waiting period. Within this time window, the access control system sends an activation command to the local auxiliary data acquisition module. The local auxiliary data acquisition module responds to the command, collecting auxiliary information such as the RFID card signal, mobile device Bluetooth broadcast messages, or voiceprint audio from the person passing through, formatting it into a unified data structure, and outputting it as auxiliary identification data. This step obtains an additional dimension of authentication from the physical environment without relying on network communication.

[0055] In some embodiments, auxiliary identification data can be acquired in multiple ways: Optionally, the access control system activates a near-field communication reader; receives an encrypted identity token sent by a smart device carried by a person; decrypts and verifies the format of the encrypted identity token to generate auxiliary identification data. Optionally, the access control system activates a directional microphone array; records the audio of a specific wake-up word spoken by the person; and performs Mel-frequency cepstral coefficient extraction on the audio to generate auxiliary identification data. It is understood that other infrared depth sensing-based methods can also be used to acquire auxiliary identification data, and this is not limited here.

[0056] In some embodiments, there may be instances where individuals place their RFID cards deep inside backpacks with electromagnetic shielding, or their smartphones' Bluetooth broadcast module is in sleep mode, resulting in the local auxiliary data acquisition module failing to capture any valid signal during the preset suspicious waiting period. In response, the access control system can trigger an interactive prompt module, guiding individuals to proactively present their auxiliary verification medium or adjust their position via voice announcement or screen display, and resetting the timer for the preset suspicious waiting period after the prompt.

[0057] S105. Calculate the reasonable distribution index of local difference feature data on the spatial location index, and perform cross-fusion verification of dynamic weights by combining the auxiliary identification data and the auxiliary registration data associated with the target candidate identifier to obtain the compensation judgment result.

[0058] The distribution rationality index refers to an assessment value used to quantify whether the spatial distribution of local difference characteristic data conforms to the laws of natural physiological changes, such as assessing whether occluded areas present continuous blocks. Dynamic weights represent weight coefficients that are adjusted in real time based on the credibility of each data point or the degree of environmental interference during the cross-fusion verification process. Cross-fusion verification represents the process of comprehensively calculating data from multiple different modalities or sources to arrive at a final conclusion. The compensation judgment result represents the final approval or rejection instruction output after multi-dimensional data comprehensive evaluation when initial comparison is questionable.

[0059] After acquiring local difference feature data and collecting auxiliary identification data during the pending period, the system enters the final access control decision scenario. Specifically, the access control management system first analyzes the geometric distribution characteristics of the local difference feature data on the spatial location index, such as calculating the clustering degree, connectivity, or symmetry of difference points, to derive a distribution rationality index. This index is used to determine whether these differences are caused by normal local occlusion (such as wearing a mask) or by malicious image stitching or forgery. Subsequently, the system evaluates the signal strength or data integrity of the current auxiliary identification data, and assigns dynamic weights to the local difference feature data and auxiliary identification data accordingly. Finally, the system inputs the distribution rationality index, auxiliary identification data, and locally stored auxiliary registration data into a cross-fusion decision model, performs weighted scoring or logical operations according to the dynamic weights, and comprehensively derives the compensation decision result, thereby achieving a high-confidence local access decision in a weak network environment.

[0060] In some embodiments, the acquisition of cross-fusion verification and compensation judgment results can be achieved in multiple ways: Optionally, the access control system uses a density clustering algorithm to calculate the number of clusters of local difference feature data to map to a distribution rationality index; calculates the matching score between auxiliary identification data and auxiliary registration data; and performs a weighted summation of the distribution rationality index and the matching score according to a preset weight allocation strategy. If the total score meets the standard, the compensation judgment result for allowing access is output. Optionally, the access control system inputs the spatial coordinates of the local difference feature data into a pre-trained graph neural network to output a distribution rationality index; verifies the consistency between auxiliary identification data and auxiliary registration data using a hash verification algorithm; constructs a decision tree model, using the distribution rationality index and consistency result as input nodes, and outputs the final compensation judgment result through tree-like logical reasoning. It is understood that other methods based on fuzzy logic reasoning engines can also be used to acquire the cross-fusion verification and compensation judgment results, which are not limited here.

[0061] In real-world access control scenarios, security threats such as identity theft (e.g., proxy clocking in) or malicious spoofing may occur. For example, when an unauthorized person uses employee A's RFID access card or Bluetooth credential to swipe, the system will determine the credential is valid (i.e., the auxiliary identification data matches the auxiliary registration data). However, the facial features of the unauthorized person captured by the camera will show significant differences in key areas such as the eyes and facial contours compared to employee A's registered face (i.e., the local difference feature data shows extremely high anomalous mutations). Another example is that although the person entering the correct access password (auxiliary data matching), they may be wearing a full-coverage helmet or mask, causing drastic changes in the extracted local facial features compared to normal registration features, which are inconsistent with normal physiological changes. To address this, the access control management system can introduce a rejection mechanism into the fusion decision model. When the anomalous index of the local difference feature data exceeds the hard security boundary, the system will disregard the auxiliary verification score and directly output a rejection compensation decision to prevent the risk of identity theft.

[0062] S106. When the compensation judgment result matches the preset release state, output the access control opening signal to control the access control to open.

[0063] The preset access control status refers to the legal status flag defined in the access control system's logic control unit that allows the physical barrier to be released. The access control opening signal refers to the electrical control level or digital command sent by the access control system's main control chip to external relays or motor drivers.

[0064] Specifically, the access control system analyzes the compensation judgment result output from the cross-validation process. When the analysis result matches the preset access state defined in the memory, the main control program of the access control system calls the hardware abstraction layer interface to write a high-level signal to the GPIO pin controlling the access lock, or sends a specific unlocking protocol data packet via the RS485 bus to generate an access control opening signal. After receiving this signal, the physical actuator of the access control system performs the unlocking action, completing the passage process.

[0065] In some embodiments, the access control unlock signal can be output in several ways: Optionally, the access control management system generates an unlock command packet containing a timestamp and device signature; the command packet is sent to the access control controller via an internal CAN bus; the access control controller verifies the signature and then drives a relay to close. Optionally, the access control management system sends an encrypted unlock feature code to the smart lock's Bluetooth module; after the smart lock decrypts and verifies the feature code, it controls a stepper motor to rotate and release the mechanical lock. It is understood that other methods based on dry contact signal control can also be used to output the access control unlock signal, and this is not limited here.

[0066] To address the issue of frequent flagging due to outdated feature data caused by the inability of local feature libraries to receive timely updates from the cloud in weak network environments, this application further introduces a local feature caching and dynamic update mechanism based on the above embodiments. The method provided in this embodiment will be described in more detail below. Please refer to... Figure 2 This is another flowchart illustrating the access control method for weak network environments in this application embodiment.

[0067] S201. Acquire the target image, perform feature matching processing on the target image and the local feature library to obtain the target candidate identifier and the corresponding initial matching score.

[0068] Refer to step S101, which will not be repeated here.

[0069] S202. When the initial matching score falls into the preset doubtful interval, extract the current fine-grained feature map with spatial topological structure from the target acquisition image.

[0070] Refer to step S102, which will not be repeated here.

[0071] S203. Based on the baseline feature map of the target candidate identifier, perform differential analysis on the current fine-grained feature map in each spatial dimension, extract the feature components that exceed the significance threshold of difference and their corresponding spatial location indices, and obtain local differential feature data.

[0072] Refer to step S103, which will not be repeated here.

[0073] In a weak network environment, in order to make the most of the cloud's high computing power to correct the local feature database without affecting local processing efficiency, the access control management system performs the following steps.

[0074] S204. Obtain the connectivity status parameters of the current network communication link and calculate the network quality assessment index.

[0075] Among them, connectivity status parameters represent the underlying communication data statistically analyzed by the network interface layer of the access control management system, such as packet loss rate and round-trip latency. Network quality assessment indicators are a comprehensive value used to quantify the current network carrying capacity.

[0076] Specifically, the access control system periodically sends probe packets to the gateway or cloud server, analyzes the response time and number of lost packets, and obtains connectivity parameters. The access control system then inputs these parameters into a preset evaluation function to calculate network quality assessment indicators, providing a basis for subsequent data compression strategies.

[0077] In some embodiments, network quality assessment metrics can be calculated in several ways: Optionally, the access control system extracts the retransmission rate parameter of the TCP protocol stack; combines it with the average latency of ICMP probes; and calculates the network quality assessment metrics using a weighted summation formula. Optionally, the access control system monitors the actual throughput of the network interface; calculates the ratio of throughput to theoretical bandwidth; and maps this ratio to the network quality assessment metric. It is understood that other methods based on wireless signal strength indicators can also be used for calculation, and this is not limited here. In some embodiments, there may be extreme cases where network communication links are intermittently disconnected, leading to continuous timeouts of probe packets. In response, the access control system can forcibly set the network quality assessment metric to the lowest level and suspend all unnecessary network requests, retaining only basic heartbeat monitoring.

[0078] S205. Based on network quality assessment metrics, compress and encode local difference feature data to generate a lightweight feature verification package.

[0079] Compression coding refers to the process of reducing data redundancy through mathematical transformations. Lightweight feature verification packets refer to data encapsulation formats that have been reduced in size and are suitable for transmission in low-bandwidth networks.

[0080] Specifically, the access control system dynamically selects the compression algorithm strength based on network quality assessment indicators. The worse the network quality, the higher the compression ratio selected. The access control system performs quantization and Huffman coding on local difference feature data, removes redundant information, and encapsulates it into a lightweight feature verification packet to ensure successful transmission even under weak network conditions.

[0081] In some embodiments, compression encoding can be implemented in several ways: Optionally, the access control system performs discrete cosine transform on the local difference feature data; discards high-frequency coefficients; and performs run-length encoding on the low-frequency coefficients to generate a lightweight feature verification package. Optionally, the access control system employs an autoencoder network; inputs the local difference feature data into the encoder; and extracts low-dimensional latent variables from the intermediate layers as a lightweight feature verification package. It is understood that other dictionary-based encoding methods can also be used to achieve compression, which are not limited here.

[0082] In some embodiments, when network bandwidth is extremely limited, the system may use an extremely high compression ratio, causing key identity markers (such as high-frequency details like eye wrinkles and scar contours) in local difference feature data to be smoothed out, resulting in the data received in the cloud losing its verification value. To address this, the access control management system can calculate the key information entropy of the feature data before compression and set a hard upper limit on the compression ratio to ensure that the lightweight feature verification package still has basic usability for cloud verification.

[0083] In some embodiments, to adapt to data transmission in extremely weak network environments, the access control management system determines the target dimensionality reduction scale and target quantization step size based on network quality assessment indicators; based on a preset dimensionality reduction projection matrix, it performs dimensionality reduction processing on local differential feature data according to the target dimensionality reduction scale to obtain differential dimensionality reduction vectors; based on the target quantization step size, it discretizes and encodes the differential dimensionality reduction vectors, adds index information of target candidate identifiers, and encapsulates them to generate a lightweight feature verification package.

[0084] The target dimensionality reduction scale refers to the number of principal components retained or the compression ratio when reducing data dimensionality, such as reducing 256-dimensional features to 64-dimensional features. The target quantization step size represents the numerical interval used when converting continuous floating-point data into discrete integers. The dimensionality reduction projection matrix represents a pre-trained mathematical transformation matrix used to map high-dimensional features to a low-dimensional space. The difference dimensionality reduction vector represents the smaller data sequence obtained after dimensionality reduction. Discretization encoding refers to the compression process of mapping continuous feature values ​​to a finite number of discrete states to significantly reduce the data transmission volume.

[0085] This step is triggered when the system determines that the current network quality is poor, but cloud computing power is still needed to verify local difference features. Specifically, the access control management system first analyzes the current network quality assessment indicators. Based on bandwidth and latency, it adaptively searches for matching target dimensionality reduction scale and target quantization step size from a preset policy library. The worse the network quality, the larger the dimensionality reduction scale and the wider the quantization step size. Next, the system calls the dimensionality reduction projection matrix stored locally and projects the high-dimensional local difference feature data to a low-dimensional subspace through matrix multiplication, removing redundant information and generating a difference dimensionality reduction vector. Subsequently, the system truncates and discretizes the floating-point values ​​in the difference dimensionality reduction vector according to the target quantization step size, converting them into integer or bitmap formats with minimal space usage. Finally, the system appends the index information of the target candidate identifier to the encoded data, encapsulating it into a lightweight feature verification package to ensure successful data reporting even in extremely weak network environments.

[0086] In some embodiments, the lightweight feature verification package can be generated in several ways: Optionally, the access control system queries a mapping table based on network quality assessment indicators to obtain the target dimensionality reduction scale and quantization step size; it uses the dimensionality reduction projection matrix generated by principal component analysis to perform a linear transformation on the local differential feature data to obtain the differential dimensionality reduction vector; it performs uniform quantization on the differential dimensionality reduction vector and concatenates the identifier index to generate the lightweight feature verification package. Optionally, the access control system inputs the network quality assessment indicators into the adaptive control module to dynamically calculate the target dimensionality reduction scale and quantization step size; it uses a matrix obtained by the local preserving projection algorithm to perform manifold dimensionality reduction on the local differential feature data to extract the differential dimensionality reduction vector; it uses the K-Means clustering algorithm to perform non-uniform discretization encoding on the differential dimensionality reduction vector and packages the index information to generate the lightweight feature verification package. It is understood that other methods based on hash dimensionality reduction and scalar quantization can also be used to generate the lightweight feature verification package, which is not limited here. It should be noted that the dimensionality reduction projection matrix is ​​periodically updated synchronously from the cloud when the network connectivity is good to ensure the feature fidelity of the dimensionality reduction process.

[0087] S206. Send the lightweight feature verification package to the cloud server for comparison, and establish a corresponding asynchronous listening callback task to obtain the comparison verification results returned by the cloud server.

[0088] Among these, the asynchronous listener callback task represents a non-blocking program execution mechanism, allowing the main thread to continue executing local logic while waiting for network responses in the background. The comparison and verification result refers to the judgment information returned by the cloud server after processing using a high-computing-power model.

[0089] Specifically, the access control system pushes a lightweight feature verification packet into the network sending queue and registers an asynchronous listener callback task in the operating system. The main control flow of the access control system does not wait for a network response and continues to execute the local auxiliary verification logic. When the network interface receives the comparison verification result returned by the cloud server, the callback function is triggered to process the result.

[0090] In some embodiments, asynchronous listening can be implemented in several ways: Optionally, the access control management system creates an independent worker thread; initiates an asynchronous HTTP request in this thread; and triggers a callback when the request is completed via an event-driven mechanism. Optionally, the access control management system utilizes the publish / subscribe pattern of the MQTT protocol; publishes a lightweight feature verification package to a specific topic; subscribes to the result topic and processes the received messages through a callback function. It is understood that other methods based on WebSocket bidirectional communication can also be used, and this is not limited here. In some embodiments, there may be a situation where excessively long cloud response times lead to a backlog of callback tasks consuming memory. To address this, the access control management system can set an absolute lifespan for each asynchronous listening callback task; tasks that do not respond within the timeout period will be forcibly destroyed by the system, and the associated memory resources will be released.

[0091] S207. When the comparison and verification result is received through the asynchronous listening callback task, parse the cloud feature repair data carried in the comparison and verification result.

[0092] Among them, cloud-based feature repair data refers to the correction parameters generated by the cloud server after reverse deduction or feature completion of the lightweight feature verification package based on its high-precision model.

[0093] Specifically, after the callback function is triggered, the access control system unpacks and verifies the received network data packets. The system extracts the payload from the verification results and parses the cloud-based feature repair data. This data contains precise guidance for correcting blurred or degraded portions of the local baseline feature map.

[0094] In some embodiments, parsing can be implemented in multiple ways: Optionally, the access control system verifies the digital signature of the data packet; decrypts the data packet content; and extracts cloud-based feature repair data according to a predefined JSON format. Optionally, the access control system reads the protocol identifier in the data packet header; calls the corresponding deserialization function based on the identifier; and converts the binary stream into a cloud-based feature repair data structure. It is understood that other parsing methods based on the Protobuf protocol can also be used, and are not limited here. In some embodiments, network transmission errors may cause the parsed cloud-based feature repair data format to be corrupted. To address this, the access control system can perform strict cyclic redundancy checks; if the check fails, the data packet is discarded to prevent erroneous repair data from being introduced into the local feature library.

[0095] S208. Based on the cloud-based feature repair data, the baseline feature map of the target candidate identifier in the local feature library is overwritten to obtain the local baseline correction map.

[0096] Overwriting refers to replacing or updating the data at the corresponding address in the original storage medium with new data. Local benchmark correction maps refer to feature templates with higher accuracy after optimization using cloud-based data.

[0097] Specifically, the access control system locates the storage block corresponding to the target candidate identifier in the local feature library. The system then performs matrix fusion of the cloud-based feature repair data with the original baseline feature map, or directly replaces damaged dimensions, to complete the overwrite operation and generate a local baseline correction map. This step enables asynchronous enhancement of the local lightweight feature library using high-performance cloud computing.

[0098] In some embodiments, overwriting can be implemented in several ways: Optionally, the access control management system calculates the weighted average of the cloud-based feature repair data and the baseline feature map; writes the calculation result to the original storage address; and updates the index table of the local feature library. Optionally, the access control management system extracts the incremental matrix from the cloud-based feature repair data; performs matrix addition between the incremental matrix and the baseline feature map; and saves the result as a local baseline correction map. It is understood that other feature update methods based on Kalman filtering can also be used, and are not limited here.

[0099] In some embodiments, a sudden power outage during the overwrite process may cause corruption of the feature library file. To address this, the access control management system employs a dual-zone backup storage mechanism. Before overwriting, the original baseline feature map is backed up to the backup zone, and the master boot record is updated only after the overwrite operation is complete, ensuring data consistency and security.

[0100] In the local verification process, the access control management system continues to execute the multimodal fusion judgment process.

[0101] S209. During the preset waiting period for suspected cases, auxiliary identification data is obtained through the local auxiliary acquisition module of the access control management system.

[0102] Refer to step S104, which will not be repeated here.

[0103] S210. Perform cross-fusion verification on the local difference feature data, auxiliary identification data and auxiliary registration data associated with the target candidate identifier to obtain the compensation determination result.

[0104] Refer to step S105, which will not be repeated here.

[0105] S211. When the compensation judgment result matches the preset release state, output the access control opening signal to control the access control to open.

[0106] Refer to step S106, which will not be repeated here.

[0107] In some embodiments, to prevent forgery attacks that exploit local feature splicing, the access control management system extracts the spatial distribution features of local difference feature data in the target image and calculates the difference clustering degree when the compensation judgment result matches the preset release state. When the difference clustering degree is lower than a preset security threshold, an anti-counterfeiting challenge command is generated. The access control terminal's interaction module executes the anti-counterfeiting challenge command to collect liveness response features. After the liveness response features pass verification, an access control opening signal is output to control the access control to open. When the difference clustering degree is greater than or equal to the preset security threshold, an access control opening signal is output to control the access control to open.

[0108] Spatial distribution features represent the geometrical arrangement of differential features on a two-dimensional image plane. Differential clustering degree quantifies whether differential features exhibit a natural and continuous spatial distribution. Anti-counterfeiting interrogation commands refer to random control commands that require personnel to perform specific actions. Liveness response features represent the physiological response data of personnel to interrogation commands.

[0109] Specifically, the access control system analyzes the coordinates of local difference feature data and calculates the dispersion of its spatial distribution. If the difference features appear fragmented or unnaturally pieced together (low difference clustering), the system determines that there is a risk of forgery and immediately generates an anti-forgery query command (such as requiring blinking or head shaking). The interaction module guides the person to perform the action, the system collects and verifies the liveness response features, and only outputs the access control open signal after the verification is successful, thus ensuring both access efficiency and security.

[0110] In some embodiments, the calculation of difference clustering degree and liveness verification can be achieved in multiple ways: Optionally, the access control system calculates the convex hull area of ​​the coordinate points of local difference feature data; calculates the ratio of the number of feature points to the convex hull area as the difference clustering degree; and verifies the liveness response features by analyzing the changes in the optical flow field in continuous video frames. Optionally, the access control system uses the DBSCAN density clustering algorithm to calculate the number of clusters of difference features; uses the reciprocal of the number of clusters as the difference clustering degree; and verifies the liveness response features by extracting the three-dimensional displacement trajectory of facial key points. It is understood that other liveness verification methods based on near-infrared reflectance analysis can also be used, and are not limited here. In some embodiments, there may be situations where dim ambient light causes the liveness response feature acquisition to fail. To address this, the access control system will simultaneously increase the brightness of the access control terminal's supplementary light while generating anti-counterfeiting interrogation commands to ensure that the image sensor can clearly capture the micro-expressions or movement changes of personnel.

[0111] In order to enable the self-evolution of the local feature database, the access control system performs feature caching and lifecycle management after granting access.

[0112] S212. In response to the access control opening signal, perform association mapping between the local difference feature data and the target candidate identifier in the local feature library to generate a difference cache template.

[0113] In this context, association mapping represents establishing a connection between two data entities in the database. A difference cache template refers to a temporarily stored block of feature data that records minor recent changes in a person's appearance.

[0114] Specifically, after confirming that the access control system has opened the door, it determines that the current local difference feature data belongs to a legitimate feature evolution. The access control system encapsulates this data into a difference cache template and establishes a foreign key association between it and the target candidate identifier in the local feature library. This allows the system to remember temporary changes in the user, such as wearing a mask or glasses, improving the re-identification rate in the short term.

[0115] In some embodiments, the association mapping can be implemented in several ways: Optionally, the access control system inserts a new record into a relational database; stores local difference feature data in a BLOB field; or stores the target candidate identifier in a foreign key field. Optionally, the access control system appends the difference cache template to the corresponding value list in an in-memory key-value pair database, using the target candidate identifier as the key. It is understood that other graph database-based association methods can also be used, and are not limited here. In some embodiments, frequent access may cause the number of difference cache templates to surge, exhausting local storage. To address this, the access control system can set a maximum limit on the number of difference cache templates for each target candidate identifier. When the limit is reached, a first-in-first-out (FIFO) strategy is used to overwrite the oldest template.

[0116] In some embodiments, the access control management system will solidify the stable feature changes of high-frequency hits over a long period of time. That is, the access control management system will obtain multiple differential cache templates associated with the target candidate identifier in the local feature library; perform feature fusion on the multiple differential cache templates according to the cumulative hit frequency to obtain feature incremental update data; and update the baseline feature map of the target candidate identifier in the local feature library based on the feature incremental update data.

[0117] Feature fusion refers to the process of merging multiple related feature data into a more representative feature vector through mathematical operations. Feature increment update data refers to the feature compensation amount extracted after fusion, which reflects the long-term stable changes in a person's appearance.

[0118] Specifically, the access control system selects differential cache templates with high and stable cumulative hit frequencies, performs weighted average or principal component analysis on them, and extracts common feature incremental update data. The access control system then overlays this incremental data onto the original baseline feature map, completing the adaptive evolution of the local feature library, enabling the system to adapt to the natural aging or long-term changes in people's appearance.

[0119] In some embodiments, feature fusion can be implemented in several ways: Optionally, the access control system uses the cumulative hit frequency as weight; performs a weighted summation on multiple difference cache templates; and performs L2 normalization on the summation result to obtain feature incremental update data. Optionally, the access control system stacks multiple difference cache templates into a matrix; calculates the covariance of the matrix; and extracts the eigenvector corresponding to the largest eigenvalue as feature incremental update data. It is understood that other methods based on cluster center extraction can also be used, and this is not limited here.

[0120] In some embodiments, it is possible that personnel frequently wear specific protective masks to enter and exit the system on a particular day due to special operations. This could cause the system to misjudge this anomalous change as a long-term stable high-frequency feature and incorporate it into the baseline data, thereby contaminating the local feature library with non-routine features. To address this, the access control management system calculates the time span and variance between each difference cache template before performing feature fusion. If the variance is too large or the time span is concentrated within a very short period, it indicates that the feature changes are unstable, and the fusion update process is terminated to ensure the purity of the baseline feature map.

[0121] S213. Obtain the lifecycle tags of multiple differential cache templates.

[0122] Among them, the lifecycle tag represents the metadata attached to the differential cache template, which is used to record the usage status and time attributes of the template.

[0123] Specifically, the access control system periodically scans the local feature library, reads the attribute fields of all differential cache templates associated with the target candidate identifier, and extracts lifecycle tags containing the last trigger timestamp and cumulative hit frequency, providing an evaluation basis for subsequent template cleanup.

[0124] In some embodiments, lifecycle tags can be obtained in several ways: Optionally, the access control system executes an SQL query; traverses a template table with specific identifiers; and extracts the timestamp and frequency fields. Optionally, the access control system traverses a linked list structure in memory; and reads the lifecycle structure data defined at the head of each node. It is understood that other methods based on file system attributes can also be used, and are not limited here. In some embodiments, system clock drift may cause the last trigger timestamp to be inaccurate. To address this, the access control system can synchronize its time with the NTP server each time the network is connected, and use a monotonically increasing clock relative to the system boot time when recording timestamps to avoid logical errors caused by absolute time jumps.

[0125] S214. When the idle time of the target difference cache template exceeds the preset aging threshold or the cumulative hit frequency is lower than the preset active threshold, the target difference cache template is removed from the local feature library.

[0126] The idle time represents the difference between the current time and the last triggered timestamp. The preset aging threshold and preset active threshold are numerical standards set by the system to determine whether a template has become invalid.

[0127] Specifically, the access control management system calculates the idle time of each template and compares it with a preset aging threshold; at the same time, it checks whether the cumulative hit frequency is lower than a preset active threshold. If any condition is met, the target difference cache template is determined to be an occasional or expired feature change, and the access control management system performs a deletion operation to release storage space and maintain the efficient operation of the local feature database.

[0128] In some embodiments, removal can be implemented in several ways: Optionally, the access control management system sends a DELETE statement to the database engine; deletes records that match the conditions; and performs storage defragmentation. Optionally, the access control management system marks the memory block containing the target difference cache template as free in the memory pool; and returns its pointer to the memory allocator. It is understood that other methods based on soft deletion marking can also be used, and are not limited here. In some embodiments, batch removal operations may cause a sudden surge in system I / O load, affecting normal access control recognition. To address this, the access control management system can place the removal operations in a low-priority background queue and execute the deletion instructions in batches when the system is idle.

[0129] In this embodiment, by performing difference analysis by extracting current feature data and benchmark feature maps, and performing cross-fusion verification by combining auxiliary recognition data obtained by the local auxiliary acquisition module, the access control management system can complete high-confidence local permission determination without relying on the cloud in a weak network environment. This solves the problem of a large number of false rejections caused by the lag in cloud comparison results in a weak network environment, which leads to the system relying entirely on low-precision local algorithms. This achieves high passage efficiency and high recognition accuracy in a weak network environment.

[0130] The access control management system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of an access control management system in an embodiment of this application.

[0131] It should be noted that, Figure 3 The structure of the access control management system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0132] like Figure 3As shown, the access control management system includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in ROM 302 or a program loaded into RAM 303 from storage section 308, such as executing the methods described in the above embodiments. RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0133] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including hard disks, etc.; and communication section 309 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0134] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0136] Specifically, the access control management system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the access control method for weak network environments provided in the above embodiment.

[0137] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the access control management system described in the above embodiments; or it may exist independently and not be assembled into the access control management system. The storage medium carries one or more computer programs, which, when executed by a processor of the access control management system, cause the access control management system to implement the weak network environment access control method provided in the above embodiments.

[0138] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0139] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

Claims

1. A method for access control in a weak network environment, characterized in that, Applied to an access control management system, the method includes: Acquire a target image, perform feature matching processing on the target image and the local feature library to obtain target candidate identifiers and corresponding initial matching scores; When the initial matching score falls into the preset doubtful interval, the current fine-grained feature map with spatial topological structure in the target acquired image is extracted; Based on the baseline feature map of the target candidate identifier, perform differential analysis on the current fine-grained feature map in each spatial dimension, extract feature components that exceed the differential significance threshold and their corresponding spatial location indices, and obtain local differential feature data; During the preset waiting period for suspected cases, auxiliary identification data is obtained through the local auxiliary data acquisition module of the access control management system; Calculate the reasonable distribution index of the local difference feature data on the spatial location index, and perform dynamic weight cross-fusion verification by combining the auxiliary identification data and the auxiliary registration data associated with the target candidate identifier to obtain the compensation determination result; When the compensation determination result matches the preset release state, an access control opening signal is output to control the access control to open.

2. The method according to claim 1, characterized in that, After the step of outputting an access control opening signal to control the access control to open when the compensation determination result matches the preset release state, the method further includes: In response to the access control opening signal, the local difference feature data is associated and mapped with the target candidate identifier in the local feature library to generate a difference cache template; Obtain lifecycle tags for multiple differential cache templates; the lifecycle tags include the last trigger timestamp and / or the cumulative hit frequency; When the idle time of the target difference cache template exceeds a preset aging threshold or the cumulative hit frequency is lower than a preset active threshold, the target difference cache template is removed from the local feature library.

3. The method according to claim 2, characterized in that, After the step of performing an association mapping between the local difference feature data and the target candidate identifier in the local feature library in response to the access control opening signal to generate a difference cache template, the method further includes: Obtain multiple differential cache templates associated with the target candidate identifier from the local feature library; Based on the cumulative hit frequency, feature fusion is performed on the multiple differential cache templates to obtain feature incremental update data; Based on the feature increment update data, the baseline feature map of the target candidate identifier in the local feature library is updated.

4. The method according to claim 1, characterized in that, After the step of performing differential analysis on the current fine-grained feature map dimensionally based on the benchmark feature map of the target candidate identifier, extracting feature components exceeding the significance threshold and their corresponding spatial location indices to obtain local differential feature data, the method further includes: Obtain the connectivity status parameters of the current network communication link and calculate the network quality assessment index; Based on the network quality assessment metrics, the local difference feature data is compressed and encoded to generate a lightweight feature verification package; The lightweight feature verification package is sent to the cloud server for comparison, and a corresponding asynchronous listening callback task is established to obtain the comparison verification result returned by the cloud server.

5. The method according to claim 4, characterized in that, After the steps of sending the lightweight feature verification package to the cloud server for comparison and establishing a corresponding asynchronous listening callback task to obtain the comparison verification result returned by the cloud server, the method further includes: When the comparison and verification result is received through the asynchronous listening callback task, the cloud feature repair data carried in the comparison and verification result is parsed. Based on the cloud-based feature repair data, the baseline feature map of the target candidate identifier in the local feature library is overwritten to obtain the local baseline correction map.

6. The method according to claim 4, characterized in that, The step of compressing and encoding the local difference feature data based on the network quality assessment index to generate a lightweight feature verification package specifically includes: The target dimensionality reduction scale and target quantization step size are determined based on the network quality assessment indicators. Based on the preset dimension reduction projection matrix, the local difference feature data is subjected to dimension reduction processing according to the target dimension reduction scale to obtain the difference dimension reduction vector; The difference dimensionality reduction vector is discretized and encoded based on the target quantization step size, and the index information of the target candidate identifier is appended to generate the lightweight feature verification package.

7. The method according to claim 1, characterized in that, The step of outputting an access control opening signal to control the access control to open when the compensation determination result matches the preset release state specifically includes: When the compensation determination result matches the preset release state, the spatial distribution characteristics of the local difference feature data in the target acquisition image are extracted, and the difference clustering degree is calculated. When the degree of difference aggregation is lower than a preset security threshold, an anti-counterfeiting query command is generated; The interactive module of the access control terminal executes the anti-counterfeiting query command to collect liveness response features, and after the liveness response features pass the verification, it outputs an access control opening signal to control the access control to open; When the difference aggregation degree is greater than or equal to the preset security threshold, an access control opening signal is output to control the access control to open.

8. An access control management system, characterized in that, The access control management system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the access control management system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the access control management system, the access control management system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the access control management system, it causes the access control management system to perform the method as described in any one of claims 1-7.

Images