A regional access control method combining face recognition and behavior analysis

Abstract

The present invention discloses a regional access control method combining face recognition and behavior analysis, which belongs to the field of smart security, including: obtaining a high-sensitivity identification area of ​​a target area; if a user enters a high-sensitivity identification area, obtaining a user activity data set; performing time serialization processing on the user activity data set to obtain the processed user activity data set; identifying social behavior chain data according to the user activity data set, building a social behavior chain analysis model, and obtaining the user's social behavior characteristics; performing abnormal analysis on the social behavior characteristics, outputting abnormal behavior identification, and controlling the user's access rights based on the abnormal behavior identification. This application solves the technical problem that the existing regional access control only relies on face recognition and lacks abnormal analysis of user behavior. By combining face recognition and social behavior chain analysis, dynamic adjustment of user access rights is achieved, achieving the technical effect of improving the security and accuracy of regional access control.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security, and in particular to a regional access control method combining face recognition and behavior analysis. Background Art

[0002] In the prior art, regional access control usually uses face recognition technology to identify and verify user identities, and controls user access to specific areas according to pre-set permissions. However, relying solely on face recognition technology has certain limitations and security risks. First, although face recognition technology can identify the identity of users, it cannot effectively analyze and judge user behavior. Some suspicious or abnormal user behaviors, such as stalking, wandering, and illegal intrusion, are difficult to detect and prevent through face recognition alone. This affects the security and reliability of regional access control. Secondly, traditional regional access control methods usually use static permission settings, that is, pre-assigning access rights according to the identity of the user. This method lacks flexibility and dynamic adjustment capabilities, and cannot cope with real-time changing security needs. When the user's behavior is abnormal or there is a potential security threat, it is difficult to quickly revoke or adjust the granted access rights, which poses certain security risks. Summary of the invention

[0003] This application provides a regional access control method that combines face recognition and behavior analysis, aiming to solve the technical problem that existing regional access control relies only on face recognition and lacks analysis of abnormal user behavior.

[0004] The present application discloses a method for regional access control combining face recognition and behavior analysis, comprising: obtaining a highly sensitive identification class area of ​​a target area, wherein the access level of the highly sensitive identification class area is greater than or equal to a preset access level; if a user enters the highly sensitive identification class area, obtaining a user activity data set through a face recognition system; performing time serialization processing on the user activity data set to obtain a processed user activity data set; identifying social behavior chain data based on the processed user activity data set, constructing a social behavior chain analysis model, and obtaining the user's social behavior characteristics based on the social behavior chain analysis model; performing abnormal analysis on the social behavior characteristics, outputting an abnormal behavior identifier, and controlling the user's access rights based on the abnormal behavior identifier.

[0005] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0006] Due to the adoption of a highly sensitive identification area for acquiring the target area, wherein the access level of the highly sensitive identification area is greater than or equal to the preset access level, targeted key monitoring and management can be carried out through the division and identification of the highly sensitive area to ensure the safety of the area; when the user enters the highly sensitive identification area, the user activity data set is acquired through the face recognition system, the face recognition system can accurately identify the user's identity, and acquire the activity data set including the user's identity information and behavior data, providing a data basis for subsequent behavior analysis; in order to facilitate subsequent analysis, the acquired user activity data set is time-serialized to obtain the processed user activity data set, and the time-serialization processing arranges the user activity data in chronological order, which is convenient for analyzing the time characteristics and changing trends of the user behavior; next, according to the processed user activity data set, the user's social behavior chain data is identified, and a social behavior chain analysis model is constructed. Chain data reflects the social interactions and behavior patterns of users in the area. By building an analysis model, the social behavior characteristics of users can be extracted and quantified. Based on the social behavior chain analysis model, the social behavior characteristics of users are analyzed for anomalies and behavior anomaly identification is output. Anomaly analysis can discover abnormal patterns or potential threats in user behavior, generate behavior anomaly identification, and provide a basis for access permission control. According to the behavior anomaly identification, the user's access rights are controlled. When abnormal user behavior is detected, corresponding measures can be taken in time, such as restricting or revoking the user's access rights, dynamically adjusting the access control strategy, and ensuring the technical solution of regional security. This solves the technical problem that the existing regional access control only relies on face recognition and lacks analysis of user behavior anomalies. By combining face recognition and social behavior chain analysis, dynamic adjustment of user access rights is realized, achieving the technical effect of improving the security and accuracy of regional access control.

[0007] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a regional access control method combining face recognition and behavior analysis is provided for an embodiment of the present application;

[0009] Figure 2 A flow chart of identifying social behavior chain data in a regional access control method combining face recognition and behavior analysis is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0010] The overall idea of ​​the technical solution provided by this application is as follows:

[0011] An embodiment of the present application provides a method for regional access control that combines face recognition and behavior analysis. First, highly sensitive identification areas of the target area are obtained, and these areas whose access levels are greater than or equal to the preset access level are monitored in a focused manner. When a user enters a highly sensitive identification area, a user activity data set is obtained through a face recognition system, including user identity information and behavior data. To facilitate analysis, the acquired user activity data set is time-serialized. Next, based on the processed user activity data set, the user's social behavior chain data is identified, a social behavior chain analysis model is constructed, and the user's social behavior features are extracted. Based on the model, the user's social behavior features are analyzed for abnormalities, and a behavior abnormality identification is output. Finally, based on the behavior abnormality identification, the user's access rights are dynamically controlled, and the access rights of abnormal users are promptly restricted or revoked to ensure regional security.

[0012] Through the above scheme, an intelligent and dynamic regional access control method is realized, which makes full use of face recognition technology to accurately identify user identities, combines social behavior chain analysis, monitors and analyzes user behaviors in the area in real time, detects abnormal situations in time, and dynamically adjusts user access rights according to abnormal identification. Compared with the existing technology, it not only realizes reliable identification of user identities, but also can effectively analyze user behaviors and detect abnormalities, thus improving the security and accuracy of regional access control.

[0013] After introducing the basic principles of the present application, the non-limiting implementation methods of the present application will be specifically introduced below in conjunction with the drawings in the specification.

[0014] like Figure 1 As shown, the embodiment of the present application provides a method for area access control combining face recognition and behavior analysis, the method comprising:

[0015] S1: Acquire a high-sensitivity identification area of ​​a target area, wherein an access level of the high-sensitivity identification area is greater than or equal to a preset access level.

[0016] Specifically, first, the target area is divided, and some important or confidential areas are defined as highly sensitive identification areas according to the security and confidentiality requirements of the area. The division is based on comprehensive consideration of factors such as the importance of the area, the confidentiality of the information involved, and the frequency of entry and exit of personnel. After the division is completed, the location information of these highly sensitive identification areas is recorded and identified, for example, using a coordinate system. At the same time, a preset access level is set to define the lower limit of the access level of the highly sensitive identification area.

[0017] When a user enters the target area, the user's location coordinates are obtained in real time and compared with the coordinate information of the highly sensitive identification area to determine whether the user has entered a highly sensitive identification area. If the access level of the area where the user is located is greater than or equal to the preset access level, it means that the user has entered a highly sensitive identification area and needs to be more strictly monitored and managed.

[0018] Acquiring highly sensitive identification areas is the premise and basis for subsequent access control measures. Only by accurately identifying highly sensitive identification areas can we implement key monitoring of users entering highly sensitive identification areas, thereby effectively improving the security prevention capabilities of the entire target area.

[0019] S2: If the user enters the highly sensitive identification area, the user activity data set is obtained through the face recognition system.

[0020] Specifically, first, a face recognition system is deployed in the target area. The system uses face recognition technology and equipment, such as face recognition cameras and face recognition gates. These devices can perform real-time face collection and recognition of users entering the target area. When a user enters a highly sensitive identification area, the face recognition system is automatically triggered to start collecting facial information of the user. During the collection process, the face recognition system will not only capture and record the user's facial image, but also obtain a series of activity data of the user in the area, such as the user's movement trajectory, residence time, interactive behavior, etc. These data will be uploaded and stored in real time to form a user activity data set of the user in the highly sensitive identification area, providing the original data basis for subsequent user behavior analysis and abnormality identification.

[0021] The face recognition system automatically collects data to avoid the inefficiency and omissions of manual operations and ensure the integrity and accuracy of the data. At the same time, the collection process is carried out without the user's perception, which will not affect the user's normal activities and improve practicality and convenience.

[0022] S3: Perform time serialization processing on the user activity dataset to obtain a processed user activity dataset.

[0023] Specifically, in practical applications, since the user activity data set collected by the face recognition system is stored in the form of discrete events, and the data format of different users may be different, it is necessary to further process these raw data into time series. The purpose of time series processing is to arrange and integrate discrete activity data in chronological order to form a continuous, uniformly formatted time series data set as the processed user activity data set.

[0024] Specifically, the first step is data cleaning, which involves denoising and outlier processing of user activity data sets, and removing incomplete, inaccurate or duplicate data records. Then comes timestamp alignment. According to the timestamp information of the activity data, the data collected from different users and at different times are aligned on a unified time axis to ensure the continuity and integrity of the data. For cases where data is missing in certain time periods, interpolation algorithms can be used to complete the data to ensure the continuity of the time series, such as linear interpolation, spline interpolation, etc. Next comes data normalization. Since the numerical ranges of different types of activity data (such as location coordinates, residence time, etc.) vary greatly, these data are normalized and mapped to the same scale to facilitate subsequent analysis and modeling.

[0025] After time series processing, the original discrete and formatted user activity data sets are converted into structured and continuous time series data sets. As the processed user activity data sets, not only the data format is standardized and the data availability is improved, but also the subsequent user behavior analysis is facilitated. At the same time, through data cleaning and preprocessing, the data quality is improved, the interference of noise and outliers is reduced, and the foundation is laid for improving the accuracy of behavior analysis.

[0026] S4: Identify social behavior chain data according to the processed user activity data set, build a social behavior chain analysis model, and obtain the user's social behavior characteristics according to the social behavior chain analysis model.

[0027] Specifically, users' activities in highly sensitive identification areas often have certain social attributes, and may involve social behaviors such as walking together, talking, and interacting. They contain rich user behavior patterns and interpersonal relationship information, which can be used to analyze user behavior characteristics and abnormal behaviors.

[0028] In order to mine these social behavior data, the behavior chain analysis method is adopted. First, by deeply analyzing the user activity data after time series processing, the social behavior characteristics of different users, such as simultaneous appearance, close location, and similar activity trajectory, are identified, and the social behavior chain data between users is constructed accordingly. The social behavior chain is essentially a directed weighted graph between users. The nodes represent users, the edges represent social behaviors between users, and the weights of the edges represent the intensity or frequency of social behaviors. After obtaining the social behavior chain data, a social behavior chain analysis model is further constructed. The model adopts the architecture of graph neural network. By end-to-end learning of the topological structure and edge weight information of the social behavior chain, the social behavior characteristics of users, such as social activity, social influence, and community belonging, are automatically extracted. These characteristics comprehensively characterize the social behavior patterns of users in highly sensitive identification areas, providing an important basis for subsequent abnormal behavior analysis.

[0029] Compared with traditional social network analysis methods, the social behavior chain analysis model makes full use of the dynamic trajectory data of users in physical space, builds social relationships through spatiotemporal interaction behaviors, and can more accurately reflect the real social patterns of users in specific scenarios. At the same time, the powerful representation learning ability of graph neural networks enables the model to automatically extract complex social behavior features, improving the efficiency and accuracy of analysis.

[0030] S5: Perform an abnormal analysis on the social behavior feature, output an abnormal behavior identifier, and control the user's access rights based on the abnormal behavior identifier.

[0031] Specifically, in actual applications, not all social behavior patterns are normal and reasonable, and some abnormal or extreme behavior patterns may indicate potential violation risks. For example, interacting with others too frequently, frequently going to the edge of highly sensitive areas, and contacting a large number of different people in a short period of time are all abnormal social behaviors.

[0032] After obtaining the social behavior characteristics, an abnormal analysis is performed on them to identify these abnormal social behaviors. For example, an unsupervised learning anomaly detection algorithm is used to establish a baseline model of normal social behavior and identify abnormal behavior instances that deviate from the model. In addition, a rule-based anomaly discrimination method can be introduced to set predefined abnormal behavior rules (such as the frequency of contact with strangers exceeds a threshold), match the user's social behavior characteristics with rules, and identify abnormal behaviors that violate the rules. The results of the anomaly analysis will be output in the form of an abnormal behavior mark. The mark can be a binary variable (normal / abnormal) or an abnormal score that indicates the severity of the abnormal social behavior. Based on the abnormal behavior mark, the user's access rights in the highly sensitive mark category area are dynamically adjusted. For example, for users whose behavior anomaly marks are abnormal or whose anomaly scores exceed the warning value, their access rights level is automatically reduced to restrict them from entering certain areas with higher security levels; while for users who have been performing normally for a long time, their access rights level can be appropriately increased to improve work efficiency.

[0033] By implementing real-time monitoring and dynamic management of user social behaviors based on abnormal behavior identification, we can timely discover and control violation risks and improve the security prevention capabilities of highly sensitive areas. At the same time, through the permission adjustment mechanism driven by abnormal behavior identification, we avoid the simple and crude management method of treating everyone equally, and improve the flexibility and humanization of management while ensuring safety.

[0034] Furthermore, the embodiment of the present application also includes:

[0035] If a user enters a low-sensitivity identification area, and the access level of the low-sensitivity identification area is less than the preset access level; the user's facial information is collected through the face recognition system to obtain the user's facial image; user identity authentication is performed based on the user's facial image, and if the identity authentication is passed, a permission opening instruction is obtained, and the entry permission to the low-sensitivity identification area is opened according to the permission opening instruction.

[0036] In a feasible implementation, in addition to the highly sensitive identification area, there are some low-sensitivity identification areas in the target area, and the access level of these areas is lower than the preset access level. For these areas, it would be too cumbersome and inefficient to adopt the same strict access control method as the highly sensitive identification area, so a relatively simplified access control process is adopted.

[0037] When a user enters a low-sensitivity identification area, the face recognition system deployed in the area automatically starts to collect the user's face image information and obtain the user's identity information. Subsequently, the collected user face image is compared with the pre-stored employee face information database to verify whether the user's identity is an authorized legal identity. Among them, the identity authentication process uses mature face recognition algorithms, such as convolutional neural networks and support vector machines. If the user's identity authentication passes, it means that he is a legally authorized user, and a permission opening instruction is issued to allow the user to enter the low-sensitivity identification area; on the contrary, if the identity authentication fails, the user's entry request is rejected to ensure the access security of the area. Compared with the complex access control process of the high-sensitivity identification area, the access control of the low-sensitivity identification area relies on the one-time verification of the user's identity and does not involve continuous monitoring and analysis of user behavior, so the process is relatively simple. This hierarchical access control method can improve the efficiency of personnel entry and exit and save management costs while ensuring overall security.

[0038] Further, such as Figure 2 As shown, the embodiment of the present application also includes:

[0039] According to the processed user activity data set, the users who enter the highly sensitive identification class area and their peer users are identified, and a social network graph is constructed, wherein the network nodes in the social network graph represent the entering users, and the edges represent the peer relationships between users; the peer feature data of each user and other users are calculated, wherein the peer feature data includes the number of peers, the duration of peer participation, and the frequency of interaction between users; according to the peer feature data, a weighted network graph is generated; according to the weighted network graph, the social network graph is weightedly constructed, and social behavior chain data is output.

[0040] In a preferred embodiment, when identifying social behavior chain data according to the processed user activity data set, first, all users who enter the highly sensitive identification class area within a certain time period are identified according to the user activity data set after time serialization processing, and it is determined whether there is a peer relationship among these users. The so-called peer relationship refers to two or more users who enter the same area in a similar time period and maintain a certain proximity in space and act together. Based on the peer relationship between users, a social network graph is constructed, in which the nodes of the graph represent each user who enters the highly sensitive identification class area, and the edges between the nodes represent the peer relationship between users. Subsequently, the peer feature data between each group of peer users are further calculated, including the number of peers, the duration of peer participation, and the frequency of interaction between users. Among them, the number of peers refers to the frequency of multiple users entering the highly sensitive identification class area at the same time within a certain time range; the duration of peer participation refers to the duration of each peer of multiple users; the frequency of interaction between users refers to the frequency of communication and interaction between multiple users during the peer process, which is indirectly inferred by analyzing the changes in the proximity of multiple users in space. These peer feature data depict the closeness of social behavior between users from different aspects.

[0041] Next, a weighted network graph is generated based on the obtained peer feature data. Similar to the constructed social network graph, the weighted network graph is also composed of nodes and edges, but the difference is that the edges here not only indicate the presence or absence of peer relationships, but also have weight information indicating the closeness of peers. The weight of the edge is derived based on features such as the number of peers, the length of time spent together, and the frequency of interaction. Afterwards, the generated weighted network graph is integrated with the constructed social network graph, and weighted social behavior chain data is generated by attaching weight information to the edges of the social network graph, which not only reflects the social behavior relationships between users, but also reflects the strength of these relationships.

[0042] Through the above steps, the recognition and transformation from user activity data to social behavior chain data is realized, laying a data foundation for subsequent behavior analysis. This graph-based social behavior modeling method makes full use of the interactive information contained in the user's spatiotemporal trajectory data. Compared with simple time series data analysis, it can more accurately and meticulously characterize the user's social behavior patterns in highly sensitive areas.

[0043] Furthermore, the embodiment of the present application also includes:

[0044] Obtain the path similarity of two users, the overlapping stay time of the two users in the same area, and the distance between the two users; according to the path similarity of the two users, the overlapping stay time of the two users in the same area, and the distance between the two users, obtain the path peer index, the time peer index, and the distance peer index; if the path peer index, the overlapping time peer index, and the distance peer index are all greater than a preset threshold, identify them as peer users.

[0045] In a preferred embodiment, users with peer relationships are identified from all users entering the highly sensitive identification class area. Taking two users as an example, first, the movement trajectory data of the two users in the highly sensitive identification class area is extracted, and the path similarity of the two users is calculated. Path similarity reflects the consistency of the movement trend of the two users in space, and is measured by calculating indicators such as the distance or angle between the two trajectories. At the same time, the overlapping residence time of the two users in the same area is calculated, that is, the length of time the two users stay in a specific area at the same time. In addition, the spatial distance between the two users is continuously tracked, and statistics such as the average distance and the minimum distance of the two users during the entire activity process are calculated. These indicators reflect the proximity of the two users in time and space from different angles. Afterwards, the calculated path similarity, the overlapping residence time of the two users in the same area, and the distance between the two users are further processed to generate a higher level of peer indicators. Specifically, the path similarity is mapped into a path peer indicator, which indicates the similarity of the movement trajectories of the two users; the overlapping stay time of the two users in the same area is mapped into a time peer indicator, which indicates the degree of synchronization of the two users in the time dimension; the distance between the two users is mapped into a distance peer indicator, which indicates the closeness of the two users in the spatial dimension. These peer indicators are usually obtained by normalizing or thresholding the original indicators for unified measurement and comparison. Then, the generated path peer indicator, time peer indicator and distance peer indicator are compared with the corresponding preset thresholds. Only when all three indicators exceed the corresponding thresholds, the two users are identified as peer users. Among them, the preset threshold is set according to experience, and can also be automatically learned through methods such as data analysis and model training. Through the joint judgment of multiple peer indicators, the accuracy of peer user identification is effectively improved and the misjudgment rate is reduced.

[0046] Furthermore, the embodiment of the present application also includes:

[0047] Initialize the graph neural network; construct a feature vector sample for each user according to the social behavior chain data, including a user social centrality sample and a user interaction frequency change sample; learn and train to obtain a social behavior chain analysis model according to the user social centrality sample and the user interaction frequency change sample, as well as identification information that identifies abnormal behavior.

[0048] In a feasible implementation, in order to obtain a social behavior chain analysis model, first, the graph neural network is initialized. The graph neural network is a deep learning model specifically used to process graph structure data. By performing end-to-end joint learning on the attribute information of the nodes and the structural information of the edges in the graph, the characteristic patterns contained in the graph data can be efficiently mined. The initialization process includes determining the number of layers of the graph neural network, the number of neurons in each layer, the selection of activation functions, the weight initialization method and other hyperparameters. Subsequently, a feature vector sample is constructed for each user based on the social behavior chain data. By analyzing the social behavior chain data, a feature vector is constructed for each node (i.e., each user) in the graph to characterize the social behavior characteristics of the user. These feature vector samples mainly include two categories: one is the user social centrality sample, which reflects the importance and influence of the user in the entire social network, and is obtained by calculating the user's degree centrality, betweenness centrality, proximity centrality and other indicators; the other is the user interaction frequency change sample, which reflects the dynamic change trend of the user's interaction frequency with other users, and is obtained by counting the frequency of user interaction behaviors in different time windows. These two types of samples characterize the role and status of the user in the social behavior chain from both static and dynamic perspectives. Afterwards, the constructed user social centrality samples and user interaction frequency change samples, as well as some pre-labeled user behavior abnormality identification information, are used to train the initialized graph neural network to obtain a social behavior chain analysis model. During the training process, the graph neural network continuously adjusts the model parameters to learn how to map the user's feature vector to its abnormal behavior identification, thereby realizing automatic analysis and anomaly detection of the user's social behavior. The trained social behavior chain analysis model can be regarded as a discriminator of social behavior anomalies, which can automatically identify suspicious abnormal patterns from the user's social behavior data.

[0049] By utilizing the powerful graph data processing capabilities of graph neural networks and end-to-end representation learning, deep features that characterize user social behavior are automatically extracted, and these features are further mapped to behavioral anomaly markers to achieve intelligent and automated social behavior anomaly detection. Compared with traditional rule-based or statistical anomaly detection methods, this method does not require manual design of features and discrimination rules, and has better adaptability and generalization capabilities.

[0050] Furthermore, the embodiment of the present application also includes:

[0051] Record the user's current access permission level; obtain the permission reduction step size according to the abnormal behavior identifier, and reduce the user's current access permission level with the permission reduction step size.

[0052] In a preferred embodiment, before adjusting the user's access rights, the user's current access rights level is first obtained and recorded, and the user's current access rights level is obtained from the database of the rights management system. The user's current access rights level reflects the area range and confidentiality of the data resources that the user can access, and is divided into multiple levels, which represent the size of the access rights from low to high. For example, the access rights level can be divided into "public level", "internal level", "confidential level", "top secret level", etc., and each level corresponds to a different area access and data access range. Subsequently, according to the output abnormal behavior identification, the corresponding permission reduction step is obtained, and the user's current access rights level is reduced with the permission reduction step. Among them, the permission reduction step indicates the amplitude or level by which the user's access rights level should be reduced when the user's behavior is found to be abnormal. The permission reduction step is dynamically set according to the severity of the abnormal behavior. For example, for minor abnormal behavior, the permission is reduced by one level; for serious abnormal behavior, the permission is reduced by multiple levels, or even the user's access rights are directly reduced to the lowest level. Through the dynamic permission adjustment strategy based on abnormal behavior, the access ability of abnormal users can be timely and effectively restricted, and their threats to sensitive areas and data resources can be reduced.

[0053] By tracking abnormal user behavior in real time and dynamically adjusting their access rights level, we can accurately and efficiently implement access control for abnormal users and improve our ability to deal with internal threats. At the same time, differentiated permission control strategies for areas with different security levels ensure the security of highly sensitive areas while also taking into account the protection of low-sensitivity areas, providing a comprehensive, multi-level access security protection solution.

[0054] Furthermore, the embodiment of the present application also includes:

[0055] Identify the user's historical behavior abnormal identification information; perform abnormal continuity analysis based on the historical behavior abnormal identification information to obtain an abnormal continuity index; establish a mapping relationship between the abnormal continuity index and the permission reduction step length, and perform feedback optimization on the permission reduction step length based on the mapping relationship.

[0056] In a preferred implementation, in order to improve the accuracy and adaptability of access rights control, a feedback optimization mechanism based on historical behavior abnormality identification is introduced.

[0057] First, identify the user's historical behavior abnormal identification information, obtain the user's behavior abnormality record in the past period of time, and provide a data basis for the subsequent abnormal continuity analysis. In specific implementation, by querying the behavior abnormal identification information database, all historical abnormal identification data related to the current user are retrieved and arranged in chronological order. These historical abnormal identification information include data in multiple dimensions such as the type, occurrence time, and severity of abnormal behavior. Then, based on the identified historical behavior abnormal identification information, perform abnormal continuity analysis to obtain an abnormal continuity index that reflects the persistence of the user's abnormal behavior. The abnormal continuity index is used to measure the frequency and regularity of the user's abnormal behavior in a period of time, and comprehensively considers factors such as the number of occurrences, time intervals, and severity trends of abnormal behaviors. The calculation of the abnormal continuity index adopts the sliding time window method. For example, the number of abnormal behaviors of the user in the past n consecutive days is counted, and the abnormal continuity index is divided into multiple levels according to the number of times. The higher the abnormal continuity index, the more frequent and continuous the user's abnormal behavior, and the greater the potential risk. Subsequently, a mapping relationship between the abnormal continuity index and the permission reduction step is established, and the original permission reduction step is dynamically optimized based on the mapping relationship. The mapping relationship can be in various forms such as piecewise function and decision tree. Among them, the higher the abnormal continuity index, the larger the permission reduction step should be, that is, stricter permission control should be carried out for users who frequently and continuously exhibit abnormal behaviors. Through this mapping relationship, during the permission reduction process, the size of the permission reduction step is dynamically adjusted to match the user's actual abnormal risk level. This adjustment can prevent the permission reduction from being too radical or too conservative, thereby achieving a balance between controlling abnormal risks and ensuring normal access.

[0058] Furthermore, the embodiment of the present application also includes:

[0059] If the user's facial image is blocked, obtain peer users; obtain peer user facial images according to the social behavior chain analysis model, and identify the user in the peer user facial images.

[0060] In a preferred embodiment, in the process of collecting user face information and obtaining user face images through a face recognition system, the user face may be blocked and cannot be directly recognized. In order to solve this problem, first, it is determined whether the collected user face image is blocked, for example, the user wears items such as masks, sunglasses, hats, or the facial features are incomplete due to factors such as ambient lighting and shooting angles. In order to determine whether the face image is blocked, a key point detection-based method is used to evaluate the quality of the face image by analyzing the visibility and integrity of key facial feature points (such as eyes, nose, mouth, etc.). If the key point is seriously missing or the confidence is very low, it can be determined that the face image is blocked. When it is detected that the user's face image is blocked, the user's peers are further obtained to provide support for subsequent cross-user face recognition.

[0061] Then, the constructed social behavior chain analysis model is used to identify other users who have a peer relationship with the current user, and the facial image data of these peer users are obtained to obtain the facial images of the peer users. Specifically, the position of the current user in the social behavior chain network is used as input, and the social behavior chain analysis model is used for reasoning to find some adjacent nodes that are most closely connected to the user and interact most frequently as peer users. Then, the facial image data of these peer user candidates in a similar time period are retrieved from the historical records of the face recognition system. Next, the identity of the current user is searched in the facial images of these peer users using the face recognition algorithm. Since the activity trajectory and interaction mode of the peer user are highly similar to that of the current user, its facial image is likely to contain the identity information of the current user, so the success rate of cross-user face recognition can be greatly improved. Through the collaborative face recognition method, even if the facial image quality of the current user is poor, a more reliable identity confirmation can be achieved through the information of the peer users.

[0062] By utilizing social behavior chain analysis, exploring peer relationships between users, and fully tapping into the correlation between different user data, we can effectively overcome obstacles to face recognition. This collaborative data processing paradigm significantly improves the robustness of face recognition and reduces over-reliance on the quality of a single piece of data.

[0063] In summary, the area access control method combining face recognition and behavior analysis provided in the embodiment of the present application has the following technical effects:

[0064] Acquire the highly sensitive identification area of ​​the target area, where the access level of the highly sensitive identification area is greater than or equal to the preset access level. By identifying and dividing the highly sensitive area, determine the area range that needs to be monitored and managed, and provide the target area for subsequent user behavior analysis and access control. If the user enters the highly sensitive identification area, obtain the user activity data set through the face recognition system. When the user enters the highly sensitive area, use the face recognition technology to accurately identify the user's identity, obtain the activity data set including the user's identity information and behavior data, and provide the data basis for subsequent behavior analysis. Perform time serialization processing on the user activity data set to obtain the processed user activity data set. By performing time serialization processing on the user activity data, the user behavior data is arranged in chronological order, which is convenient for analyzing the time characteristics and change trends of the user behavior, and preparing for the construction of the social behavior chain analysis model. According to the processed user activity data set, the social behavior chain data is identified, and the social behavior chain analysis model is constructed. According to the social behavior chain analysis model, the social behavior characteristics of the user are obtained. By identifying the social behavior chain data of the user and constructing the social behavior chain analysis model, the social interaction and behavior patterns of the user in the area can be extracted and quantified, and the social behavior characteristics reflecting the characteristics of the user behavior can be obtained, which provides a basis for abnormal behavior detection. Perform abnormal analysis on social behavior characteristics, output abnormal behavior identification, and control user access rights based on abnormal behavior identification. By performing abnormal analysis on users' social behavior characteristics, abnormal patterns or potential threats in user behavior are discovered, and abnormal behavior identification is generated. Based on the abnormal identification, the user's access rights are dynamically adjusted, and measures are taken to restrict or revoke the permissions of abnormal users, thereby achieving real-time and dynamic access control and ensuring regional security.

[0065] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0066] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these changes and variations.

Claims

1. A regional access control method combining face recognition and behavior analysis, characterized in that: The method comprises: Acquire a highly sensitive identification area of ​​the target area, wherein the access level of the highly sensitive identification area is greater than or equal to a preset access level; If the user enters the highly sensitive identification area, the user activity data set is obtained through the face recognition system; Performing time serialization processing on the user activity dataset to obtain a processed user activity dataset; Identify social behavior chain data based on the processed user activity data set, build a social behavior chain analysis model, and obtain the user's social behavior characteristics based on the social behavior chain analysis model; Performing an abnormal analysis on the social behavior characteristics, outputting an abnormal behavior identifier, and controlling the user's access rights based on the abnormal behavior identifier; Among them, identifying social behavior chain data based on the processed user activity data set includes: According to the processed user activity data set, identify the users who enter the highly sensitive identification class area and their peer users, and construct a social network graph, wherein the network nodes in the social network graph represent the entering users, and the edges represent the peer relationships between the users; Calculate the peer characteristic data of each user and other users, wherein the peer characteristic data includes the number of peers, the duration of peer participation, and the frequency of interaction between users; Generate a weighted network graph according to the peer feature data; Performing weighted construction on the social network graph according to the weighted network graph, and outputting social behavior chain data; The step of identifying the user and the users traveling with him who have entered the highly sensitive identification area includes: Obtain the path similarity of two users, the overlapping stay time of two users in the same area, and the distance between the two users; According to the path similarity of the two users, the overlapping stay time of the two users in the same area and the distance between the two users, a path peer index, a time peer index and a distance peer index are obtained; If the path peer index, overlapping time peer index, and distance peer index are all greater than a preset threshold, the user is identified as a peer user; Among them, identifying social behavior chain data based on the processed user activity data set and building a social behavior chain analysis model include: Initialize the graph neural network; Constructing a feature vector sample for each user according to the social behavior chain data, including a user social centrality sample and a user interaction frequency change sample; Based on the user social centrality samples and user interaction frequency change samples, as well as identification information identifying abnormal behaviors, learning is performed to train and obtain a social behavior chain analysis model; If the user enters a low-sensitivity identification area, and the access level of the low-sensitivity identification area is lower than the preset access level, the method includes: Collecting the user's facial information through the facial recognition system to obtain the user's facial image; Performing user identity authentication based on the user's facial image, and if the identity authentication is passed, obtaining a permission opening instruction, and opening the access permission to the low-sensitivity identification area according to the permission opening instruction; The method for controlling the user's access rights based on the abnormal behavior identifier includes: Record the user's current access permission level; According to the abnormal behavior identifier, a permission reduction step length is obtained, and the user's current access permission level is reduced according to the permission reduction step length.

2. The method according to claim 1, characterized in that The method further comprises: Identify abnormal historical behavior information of users; Performing abnormal continuity analysis based on the abnormal identification information of the historical behavior to obtain an abnormal continuity index; A mapping relationship between the abnormal continuity index and the authority reduction step length is established, and feedback optimization is performed on the authority reduction step length according to the mapping relationship.

3. The method according to claim 1, characterized in that The face recognition system is used to collect the face information of the user and obtain the face image of the user. The method further includes: If the user's face image is blocked, obtain the user who is traveling with the user; According to the social behavior chain analysis model, facial images of peer users are obtained, and users are identified in the facial images of peer users.

Images