Personnel information security screening method and system based on big data

Abstract

The invention discloses a personnel information security screening method and system based on big data, and relates to the technical field of image processing, and the method comprises the steps: obtaining to-be-processed face data in a region, screening out unregistered personnel, carrying out the matching and correlation of the images of the same unregistered personnel at different image collection points through calculating the feature similarity, and carrying out the recognition of the unregistered personnel. And connecting time points and spatial positions corresponding to the images of the same unregistered person at different image acquisition points into a spatial-temporal trajectory through a trajectory reconstruction algorithm, and pushing multi-modal feature information of the person to an adjacent preset area for the identified abnormal behavior. According to the invention, through a density clustering algorithm based on time and space constraints, feature association is carried out on personnel images collected by different cameras, a cross-region behavior track of the unregistered personnel is constructed, high-risk behaviors such as wandering are detected by using track analysis and an abnormal behavior identification technology, and early warning information is generated and sent to other regions. And the initiative and the coverage range of safety management are improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically a method and system for screening personnel information security based on big data. Background Technology

[0002] Currently, personnel information security screening is widely used in venues, communities, airports, and other scenarios requiring high-level security protection. Its core technology involves quickly identifying suspicious individuals through facial recognition and information comparison. However, traditional systems primarily rely on single facial comparison results for screening, typically completing a one-time identity verification at the entrance. This single screening method carries a high risk of false positives, cannot effectively determine a person's behavioral intentions, and in real-world scenarios, lacks dynamic monitoring of personnel movement after entrance verification, further increasing security risks.

[0003] Furthermore, traditional screening methods cannot effectively correlate and reconstruct the behavioral trajectory of the same person in multi-camera, multi-area scenarios. This results in screening being limited to local areas and unable to achieve large-scale security monitoring. In particular, it is difficult to track the complete behavioral trajectory of unregistered persons who are wandering abnormally or moving across areas. Summary of the Invention

[0004] 1) Technical problems to be solved This invention provides a method and system for personnel information security screening based on big data, which can realize multi-regional and multi-modal linkage personnel information management.

[0005] (ii) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for personnel information security screening based on big data, comprising: Within a first preset area, facial data to be processed within the area is acquired. Based on image feature extraction, the facial data is compared with a dynamically updated preset personnel information database to filter out unregistered personnel. For the unregistered personnel selected, the real-time behavior trajectory is tracked using the target tracking method, and their movement path and time sequence within the first preset area are recorded. Based on the preset time threshold and spatial conditions, it is determined whether the unregistered personnel have abnormal loitering behavior. Within the second preset area, multimodal features are extracted from multiple images of unregistered persons. For images of the same unregistered person from different image acquisition points, their feature similarity is calculated for matching and association. Specifically, the features of the person images acquired from different image acquisition points are used as nodes. Based on preset time and space constraints, a density clustering algorithm is used to cluster the different nodes to obtain clustering results. Based on the clustering results, trajectory reconstruction is performed on the nodes within each cluster; wherein, the trajectory reconstruction algorithm connects the time points and spatial locations corresponding to the same unregistered person's images from different image acquisition points into a spatiotemporal trajectory. Multiple spatiotemporal trajectories of the same unregistered person are integrated into a unified trajectory database to construct a dynamically updated personnel trajectory map. The personnel trajectory map is matched and analyzed with the set sensitive areas to screen out the intersecting paths and analyze the abnormal behavior patterns of the trajectories, including dwell time, path complexity and area access frequency. For the identified abnormal behavior, the multimodal characteristics of the person, abnormal behavior conclusions and behavior analysis information are pushed to adjacent preset areas.

[0006] Furthermore, a camera is set up in the first preset area to collect the facial data images of people in the first preset area in real time. After the facial data images are preprocessed by image enhancement and noise reduction, a pre-trained facial recognition deep learning model is used to extract the corresponding feature vectors from the images. Each facial image is mapped to a feature vector of fixed length, and its vector value represents the feature of the corresponding face. The extracted feature vectors are then normalized.

[0007] Furthermore, the preset personnel information database is a feature vector database extracted from pre-registered personnel. Based on the feature comparison algorithm, the similarity between the feature vector of the face data collected in the first preset area and the feature vector set in the preset personnel information database is calculated. According to the set similarity threshold, the corresponding face data with a similarity lower than the threshold is determined as the unregistered personnel.

[0008] Furthermore, the target tracking method is used to analyze the real-time behavioral trajectory of non-registered personnel within the first preset area, and the facial data features of non-registered personnel are used as identification features. By using multi-view mapping of cameras, the location of unregistered personnel is converted into two-dimensional plane coordinates. Combined with camera timestamps, the time points of each tracking record of the location of unregistered personnel are stored in time series form. The identification features of unregistered personnel are used to match trajectory points under multiple camera views and stitch together a complete trajectory. Based on temporal and spatial similarity, cluster analysis is performed on the trajectory points of unregistered personnel to identify whether loitering behavior exists.

[0009] Furthermore, within the second preset area, the appearance and behavioral features of non-registered personnel are extracted using target detection and segmentation technology, and the facial features, appearance features, and behavioral features of registered personnel are fused together based on an attention mechanism to generate the multimodal features; Based on multimodal features, images of the same unregistered person captured by different cameras are matched to form cross-regional associations. Each camera captures an image of an unregistered person as a node, and the node's attributes include timestamp, spatial location, and multimodal features. Set matching constraints by combining timestamps and camera spatial layout information; A similarity matrix is ​​constructed for all the nodes, and global matching is performed to associate multiple nodes of the same unregistered person.

[0010] Furthermore, based on the node matching results, the images of non-registered personnel are clustered to form cross-regional trajectories, specifically: The density threshold and minimum number of nodes are set based on the similarity between nodes and the established time and space constraints; Nodes that meet the density threshold and minimum number of nodes are clustered to form multiple clusters, each cluster representing the trajectory of the same unregistered person under multiple cameras.

[0011] Furthermore, based on the nodes contained in each cluster in the clustering results, they are sorted according to timestamps to form a time series; Between any two adjacent nodes, linear interpolation is performed based on time interval and spatial distance to generate refined trajectory points, which are then connected to form trajectory segments. Multiple trajectory segments within the same cluster are integrated, and then, for the integrated trajectory segments from multiple clusters of the same unregistered person, time series and adjacent camera position associations are used to stitch the trajectory segments into a complete spatiotemporal trajectory.

[0012] Furthermore, based on the sensitive areas set in the first and second preset areas, the constructed trajectory map of unregistered personnel is compared with the geographical location of the sensitive areas using cross-detection matching analysis based on the trajectory points and the boundaries of the sensitive areas. For detected trajectory paths that intersect with sensitive areas, the dwell time is extracted to determine whether the unregistered person has abnormal loitering behavior; By analyzing the path complexity, we can determine whether there are any abnormal detours or repetitive behaviors by unregistered personnel. By analyzing the frequency of access to their area, we can determine whether they exhibit abnormal behavior such as frequently entering and exiting sensitive areas.

[0013] Furthermore, for the abnormal behaviors identified during screening, the behavioral analysis results, along with the multimodal characteristic information of the unregistered person, are pushed to an adjacent preset area in the form of a screening report; wherein, the screening report includes: Abnormal behavior conclusion, that is, whether the unregistered person has engaged in abnormal loitering, detours, or frequent entry into sensitive areas; Behavioral analysis information, namely the results of the analysis of the unregistered person's stay time and path complexity.

[0014] A big data-based personnel information security screening system includes: The face recognition module is configured as follows: Within a first preset area, facial data to be processed within the area is acquired. Based on image feature extraction, the facial data is compared with a dynamically updated preset personnel information database to filter out unregistered personnel. The behavior recognition module is configured as follows: For the unregistered personnel selected, the real-time behavior trajectory is tracked using the target tracking method, and their movement path and time sequence within the first preset area are recorded. Based on the preset time threshold and spatial conditions, it is determined whether the unregistered personnel have abnormal loitering behavior. The regional trajectory association module is configured as follows: Within a second preset area, multimodal features are extracted from multiple images of unregistered individuals. For images of the same unregistered individual from different image acquisition points, feature similarity is calculated for matching and association. Specifically, the features of the images of individuals acquired from different image acquisition points are used as nodes. Based on preset time and space constraints, a density clustering algorithm is employed to cluster these nodes, yielding clustering results. Based on these clustering results, trajectory reconstruction is performed on the nodes within each cluster. The trajectory reconstruction algorithm connects the time points and spatial locations corresponding to the same unregistered individual images from different image acquisition points to form a spatiotemporal trajectory. The abnormal behavior push module is configured as follows: Multiple spatiotemporal trajectories of the same unregistered person are integrated into a unified trajectory database to construct a dynamically updated personnel trajectory map. The personnel trajectory map is matched and analyzed with the set sensitive areas to screen out the intersecting paths and analyze the abnormal behavior patterns of the trajectories, including dwell time, path complexity and area access frequency. For the identified abnormal behavior, the multimodal characteristics of the person, abnormal behavior conclusions and behavior analysis information are pushed to adjacent preset areas.

[0015] (iii) Beneficial effects: Compared with the prior art, this invention has the following beneficial effects: This invention combines a dynamically updated personnel information database with efficient feature indexing and comparison algorithms, and incorporates multimodal features to improve the recognition rate of unregistered personnel. It also introduces target tracking and trajectory reconstruction technology, which can record the behavioral trajectory of unregistered personnel in real time within multiple preset areas, constructing a complete spatiotemporal activity path for in-depth behavioral analysis.

[0016] By using a density clustering algorithm based on time and space constraints, feature association is performed on images of people collected by different cameras to construct the behavioral trajectory of the unregistered person across regions. Trajectory analysis and abnormal behavior recognition technologies are used to detect high-risk behaviors such as loitering and generate early warning information to be sent to other areas, thereby improving the initiative and coverage of safety management. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a personnel information security screening method based on big data, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the process of identifying non-registered personnel exhibiting abnormal behavior and pushing their information to adjacent preset areas in a personnel information security screening method based on big data provided in an embodiment of the present invention. Figure 3 A schematic diagram of a personnel information security screening system based on big data provided in an embodiment of the present invention; In the picture: 10. Face recognition module; 20. Behavior recognition module; 30. Regional trajectory association module; 40. Abnormal behavior push module. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0020] Furthermore, the terms "first" and "second" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0021] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0022] Traditional personnel information big data screening technologies typically compare specific information with existing stored information in a database. Each comparison process requires an independent retrieval and query system, and data querying and comparison are isolated, failing to form a continuous analysis workflow. Furthermore, the inventors have found that traditional screening methods usually focus on identity verification at entrances or single points, failing to track personnel behavior within internal areas. This leads to delays in detecting potential abnormal behavior, and for individuals moving across areas, it is impossible to correlate and integrate their movements across different areas, hindering the formation of a comprehensive security screening and monitoring capability for individuals exhibiting abnormal behavior.

[0023] To solve the above problems, combined with Figures 1 to 3 As shown, this invention provides a method and system for personnel information security screening based on big data. After identifying unregistered personnel, target tracking and trajectory reconstruction technologies are introduced to record the behavioral trajectories of unregistered personnel in real time within multiple preset areas, and their behavior is analyzed. Subsequently, based on a density clustering algorithm with temporal and spatial constraints, images of the unregistered personnel captured by different cameras in different areas are correlated to construct cross-regional personnel behavioral trajectories. Trajectory analysis and abnormal behavior recognition technologies are used to detect high-risk behaviors such as loitering, and early warning information is generated and sent to other areas, improving the initiative and coverage of security management.

[0024] For reference here Figure 1 First, in step S10: Within a first preset area, facial data of individuals to be processed is acquired. Based on image feature extraction, the facial data is compared with a dynamically updated preset personnel information database to filter out unregistered individuals. This can be understood as deploying high-resolution cameras within the first preset area to collect real-time facial image data of all individuals within the area. The selection of the camera location and number can be comprehensively designed based on factors such as area size, monitoring density requirements, and lighting conditions to ensure full coverage and minimize blind spots.

[0025] Considering that the quality of the acquired images may degrade due to factors such as lighting, noise, or occlusion, image preprocessing is necessary. In some embodiments, methods such as contrast stretching and histogram equalization are used to optimize the brightness and contrast of the image. Then, Gaussian filtering or bilateral filtering is used to remove noise from the image. It is important to note that key edge information must be preserved during noise removal. Finally, a pre-trained face detection model, such as the deep learning MTCNN model, is used to locate the face region. It is important to note that the preprocessed face image needs to be cropped and normalized to a fixed size for convenient subsequent uniform processing.

[0026] After preprocessing the face images, a pre-trained deep learning model is used to map each standardized face image into a fixed-length feature vector, such as 128-dimensional or 512-dimensional. Understandably, the deep learning model extracts high-dimensional features of the face through multiple convolutional and fully connected layers. These features include the position of facial key points, texture and other details, which can uniquely represent a face.

[0027] The extracted feature vectors are normalized so that their modulus in Euclidean space is 1. This process is to facilitate subsequent similarity calculations.

[0028] The similarity calculation specifically involves comparing the feature vectors collected in real time with the set of feature vectors in a preset personnel information database. In some embodiments, cosine similarity or Euclidean distance can be used to calculate the similarity between two vectors, and a similarity threshold is set. If the similarity between the face feature vector collected in real time and any feature vector in the database is lower than the threshold, the face is determined to be an unregistered person.

[0029] Regarding the aforementioned dynamically updated preset personnel information database, its dynamic updates specifically include adding, modifying, and deleting the feature vectors of registered personnel. In other words, when new personnel are registered or the features of existing personnel change, the system updates the database through the management interface to ensure real-time performance and accuracy.

[0030] The above steps demonstrate that by filtering out all unregistered individuals in real time and marking their entry time and location into the first preset area, a foundation is provided for subsequent behavioral analysis.

[0031] The aforementioned personnel screening work is now incorporated into the security management of a campus, establishing a multi-area personnel screening and behavior monitoring system to ensure the safety of the campus environment. This involves monitoring for unregistered visitors entering the campus and tracking their behavior in real time to determine if they exhibit any unusual loitering behavior. Simultaneously, the system reconstructs the activity trajectories of suspicious individuals (also unregistered) across multiple areas of the campus, and then shares the identified results with other areas for joint monitoring of the suspicious individuals.

[0032] Let's assume the main gate used to enter the campus is the first preset area. Multiple high-definition cameras are installed in this area for facial data collection, capturing real-time facial images of all personnel entering the campus and transmitting them to a backend server for processing. It's important to ensure that the cameras cover the main entrance passageway.

[0033] One of the cameras captured an unfamiliar man entering the campus gate and recorded his facial image. First, the image was preprocessed using a facial recognition module for image enhancement and noise reduction. Then, a pre-trained deep learning model for facial recognition was used to extract the man's feature vector (128 dimensions). Since the extracted feature vector did not match any of the registered personnel in the database, the man was determined to be an unregistered person and marked on the monitoring interface.

[0034] After identifying unregistered individuals, step S20 is performed: For these individuals, a target tracking method is used to track their real-time behavioral trajectory, recording their movement path and time sequence within a first preset area. Based on preset time thresholds and spatial conditions, it is determined whether the unregistered individuals exhibit abnormal loitering behavior. In this step, tracking their behavioral trajectory involves using multi-view information collected by cameras to correlate the facial features, body features, and movement trajectory of the unregistered individuals, recording their dynamic behavior.

[0035] More specifically, the system continuously tracks unregistered individuals across multiple frames, using their facial data features as a globally unique identifier to avoid confusion between different targets, i.e., between tracking different individuals.

[0036] By deploying multiple cameras within a region and fusing their perspectives to achieve visual coverage, the locations of unregistered personnel are converted into two-dimensional planar coordinates. Specifically, multi-camera calibration technology is employed, using basic geometric correction methods such as planar homography transformation to map the perspectives of different cameras to the same planar coordinate system. Furthermore, while the cameras are acquiring images, a timestamp is added to each detected trajectory point, recording the target's time-series trajectory within the region. Then, algorithms based on spatiotemporal similarity, such as the commonly used Hungarian algorithm, are used to stitch together the trajectory points captured by the multiple cameras into a complete spatiotemporal trajectory.

[0037] Finally, based on the spliced ​​trajectory, cluster analysis technology is used to identify whether unregistered personnel exhibit abnormal loitering behavior. For example, in some embodiments, a preset time threshold and spatial range are used. If the target is within time t1 to t2 and stays within a range of radius r, it is determined that the unregistered personnel exhibit loitering behavior.

[0038] In summary, it is understandable that by using target tracking and trajectory analysis technology, the real-time behavioral trajectory of unregistered personnel can be fully recorded and their movement patterns analyzed. If loitering behavior is identified, the person can be marked and an alarm can be sent subsequently, providing data support for security management.

[0039] Taking the aforementioned campus security monitoring as an example, after identifying the man as an unregistered person, it is necessary to monitor his real-time trajectory and determine whether he is loitering. Using the man's facial feature vector as the identification feature, a tracking unit (relying on a camera) tracks his activities in the campus gate area in real time. The camera collects the man's location information and timestamps, and records his trajectory points. By analyzing the trajectory points, it is found that the man repeatedly stayed in the same area for more than 10 minutes (exceeding the set behavior time threshold) and his movement path was randomly distributed. This can be interpreted as his behavior path not conforming to the normal passage path. Therefore, it can be determined that the man has abnormal loitering behavior, and he can be marked as a key monitoring target. In addition, the system can also consider issuing an immediate warning.

[0040] When the identified unregistered personnel enter the second preset area, step S30 is performed: Within the second preset area, multimodal features are extracted from multiple images of unregistered personnel. For images of the same unregistered person from different image acquisition points, their feature similarity is calculated for matching and association. Specifically, the features of personnel images acquired from different image acquisition points are used as nodes, and based on preset time and space constraints, a density clustering algorithm is used to cluster different nodes to obtain clustering results. This step can be understood as acquiring images of multiple unregistered personnel within the second preset area using cameras or other acquisition devices, extracting multimodal features from them for further analysis and association matching.

[0041] The aforementioned multimodal features specifically refer to the fusion of facial features, appearance features, and behavioral features of unregistered individuals. Among these, facial features, specifically the extracted facial feature vectors, represent the unique identity information of a person's face.

[0042] Appearance feature extraction can utilize object detection and segmentation techniques to extract the body contours and appearance features of unregistered individuals, including clothing color and texture, ultimately generating an appearance feature vector. Behavioral feature extraction can combine object tracking algorithms to extract motion trajectory parameters such as speed, acceleration, and rate of change of direction, thereby representing walking characteristics such as stride length.

[0043] Finally, the above features are fused to handle complex scenarios such as partially obscured faces and blurred appearance features. Regarding the fusion method, some embodiments may employ a dynamic fusion strategy based on an attention mechanism.

[0044] After multimodal feature extraction, it is necessary to compare the similarity of personnel features from different image acquisition points to determine whether they belong to the same unregistered person. In some embodiments, the feature similarity can be calculated using the cosine similarity method, that is, by measuring the angle between two feature vectors, their similarity value is obtained, and then a threshold is set as the judgment criterion. When the feature similarity is greater than the threshold, the two images are considered to belong to the same person, and the images that meet the condition are mapped to a node, which contains the feature vector, timestamp, spatial coordinates, and the camera ID of the acquisition camera.

[0045] Matching constraints are set by combining timestamps and camera spatial layout information; for example, the movement time of people between adjacent cameras cannot exceed a reasonable range. Then, a similarity matrix is ​​constructed for all nodes, and global matching is performed using the Hungarian algorithm or the optimal transmission algorithm to associate multiple nodes of the same unregistered person.

[0046] Based on node matching results, images of unregistered individuals are clustered to form cross-regional trajectories. Specifically, a density threshold and a minimum number of nodes are set based on the similarity between nodes and temporal and spatial constraints. Nodes that meet the density conditions are clustered to form multiple clusters, each representing the trajectory of the same unregistered individual under multiple cameras. The density condition here is that it exceeds the aforementioned set density threshold and minimum number of nodes.

[0047] Finally, based on time and space constraints, a density-based clustering algorithm is used to cluster all nodes and obtain the clustering results. In some embodiments, the time constraint is defined as the time difference between nodes in the same cluster must be within a reasonable range, and the spatial constraint is based on the camera's position layout to verify the spatial continuity of the trajectory. A density-based clustering algorithm is used for node clustering. This algorithm first identifies nodes belonging to the same cluster within a preset time range and within preset spatial ranges. Nodes that do not meet the density conditions are considered noise points.

[0048] The parameters of density clustering include distance thresholds and minimum number of points, which can be fine-tuned experimentally to suit specific scenarios. For example, if three nodes of a non-registered person have similar temporal and spatial attributes, they are clustered into the same cluster, thus laying the foundation for subsequent trajectory reconstruction.

[0049] Using the campus surveillance example again, let's consider the plaza at the main gate as the second preset area. The trajectory of a man marked as unregistered begins as soon as he enters the plaza. First, multiple images of the man are captured from multiple cameras. Multimodal features are extracted from each image, including facial features, appearance features (such as clothing color), timestamps, and spatial coordinates. All captured image features are processed as nodes. By calculating the feature similarity (e.g., cosine similarity) between nodes, matching and associating them, all nodes belonging to the same man are identified. Finally, based on temporal and spatial constraints, a density clustering algorithm is used to group these nodes into clusters, generating clustering results. Based on the clustering results, a trajectory reconstruction algorithm is used to connect the time points and spatial locations of each node, generating a complete trajectory.

[0050] After completing node clustering, step S40 is performed: based on the clustering results, trajectory reconstruction is carried out for nodes within each cluster; specifically, a trajectory reconstruction algorithm is used to connect the time points and spatial locations corresponding to the same unregistered person's images from different image acquisition points to form a spatiotemporal trajectory. The main purpose of this step is to connect the time points and spatial locations of unregistered persons at different image acquisition points to form a complete spatiotemporal trajectory.

[0051] Specifically, trajectory reconstruction algorithms can connect the two-dimensional planar coordinates and timestamps of each node within a cluster to form a complete trajectory path. More specifically, linear interpolation is used to insert intermediate trajectory points between adjacent nodes to complete the movement trajectories of unregistered personnel.

[0052] Furthermore, to improve trajectory accuracy, some embodiments may employ filtering to optimize and smooth the trajectory, eliminating noise points while preserving the overall trend of the trajectory. This not only improves trajectory accuracy but also makes the trajectory data more consistent with the actual movement paths of unregistered personnel.

[0053] Finally, S50 involves integrating multiple spatiotemporal trajectories of the same unregistered individual into a unified trajectory database to construct a dynamically updated personnel trajectory map. For identified abnormal behavior, the individual's multimodal feature information is pushed to adjacent preset areas. This can be understood as follows: after reconstructing multiple spatiotemporal trajectories of unregistered individuals, it is necessary to further integrate these trajectories into a unified trajectory database and construct a dynamically updated personnel trajectory map based on the integrated trajectory data. The goal of this process is to achieve continuous cross-regional monitoring and behavioral surveillance of unregistered individuals, providing support for identifying abnormal behavior and enabling multi-regional collaboration through information push notifications.

[0054] For multiple spatio-temporal trajectories of the same unregistered person, it is first necessary to perform normalized integration based on their multi-modal feature information. During the trajectory integration process, the feature vectors of the person, that is, the above-mentioned face features and appearance features, or directly use the multi-modal features as the association identifier, and compare the node features of each trajectory through the cosine similarity algorithm.

[0055] Suppose the similarity value of two certain trajectories exceeds the set threshold, then it is considered that they belong to the same person. The trajectory integration process sorts the time points and spatial positions of multiple similar trajectories in time series to generate a complete trajectory.

[0056] Afterwards, multiple trajectories are integrated into a unified trajectory database to achieve dynamic update and multi-region linkage analysis. It should be noted that the trajectory database needs to adopt a standardized data table design, and each record includes a trajectory ID, time series, spatial coordinates, and the multi-modal feature vector of the person. In addition, considering that the database supports real-time writing and querying, whenever a new trajectory is generated, the system will automatically update it to the database.

[0057] At the same time, by analyzing the trajectory data in the database, a dynamic trajectory map of unregistered persons can be generated. The trajectory map is a structured representation form constructed based on graph theory, where each node represents a trajectory point, and the edge represents the spatio-temporal connection between adjacent trajectory points. When constructing the trajectory map, the integrated trajectories are added to the map point by point in chronological order.

[0058] Specifically, for each trajectory point, a corresponding graph node is generated. The node attributes include the above-mentioned time stamp, coordinate position, and camera ID. Then, according to the time series of the trajectory points, adjacent nodes are connected to generate a directed edge. The edge attributes include the time difference and spatial distance. For identified abnormal behaviors, such as abnormal wandering or long-term stay, an abnormal mark is added to the corresponding node or edge for subsequent analysis.

[0059] For identified abnormal behaviors, the system will trigger an early warning mechanism and push the multi-modal feature information of the person to adjacent preset areas. The pushed content includes the person feature vector for cross-region identification, and also includes behavior descriptions, such as the time period and location of abnormal wandering, as well as the starting point, ending point positions of the trajectory, and the camera IDs passed by, etc. After receiving the pushed information, the adjacent areas will use the received feature data for matching and tracking to achieve seamless supervision across regions.

[0060] In the above example of campus security monitoring, if it is monitored that the unregistered man behaves abnormally in the square of the second preset area, then his multi-modal feature information will be pushed to the campus monitoring center and other adjacent preset areas (such as the gymnasium, laboratory building, library, etc.).

[0061] Specifically, the man's trajectory is integrated into a trajectory database to form a dynamically updated personnel trajectory map. After identifying the man's loitering behavior in the campus gate area and his frequent entry and exit from multiple public areas in the plaza area, exhibiting abnormal behavior, a push notification is generated, including: facial feature vector, behavioral description (loitering behavior and abnormal activity), and spatiotemporal trajectory data (including time series and spatial location). This push notification is sent to the monitoring systems of adjacent areas such as the gymnasium and laboratory building via a distributed message queue, prompting relevant areas to strengthen monitoring. The monitoring centers of adjacent areas activate monitoring strategies based on the push notification, identifying and tracking the man's activities in real time, so that subsequent management personnel can track and manage this unregistered and abnormally behaving individual.

[0062] In summary, it is understandable that dynamically updated trajectory maps can intuitively display the movement trajectories of unregistered personnel and support graph-based behavior analysis. The cross-regional linkage function enhances the system's intelligence level and provides strong technical support for on-site personnel management.

[0063] To implement the personnel information security screening method designed above, this invention provides a personnel information security screening system based on big data, which is described herein with reference to... Figure 3 The facial recognition module 10 is used to collect facial data and extract features within the area. By comparing with a dynamically updated personnel information database, it quickly filters out unregistered personnel. The behavior recognition module 20 tracks the filtered unregistered personnel, records their behavioral trajectories, and determines whether there is any abnormal loitering behavior based on time and space conditions.

[0064] The regional trajectory association module 30 is used to extract multimodal features of unregistered personnel, and to match and associate the images of the collection points based on the density clustering algorithm to complete the reconstruction of the spatiotemporal trajectory. Finally, there is an abnormal behavior push module 40 that integrates the trajectory data of unregistered personnel and constructs a dynamically updated trajectory map. It achieves joint supervision by identifying abnormal behavior and pushing relevant information to adjacent areas.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.

Claims

1. A method for screening personnel information security based on big data, characterized in that, include: Within a first preset area, facial data to be processed within the area is acquired. Based on image feature extraction, the facial data is compared with a dynamically updated preset personnel information database to filter out unregistered personnel. For the unregistered personnel selected, the real-time behavior trajectory is tracked using the target tracking method, and their movement path and time sequence within the first preset area are recorded. Based on the preset time threshold and spatial conditions, it is determined whether the unregistered personnel have abnormal loitering behavior. Within the second preset area, multimodal features are extracted from multiple images of unregistered persons. For images of the same unregistered person from different image acquisition points, their feature similarity is calculated for matching and association. Specifically, the features of the person images acquired from different image acquisition points are used as nodes. Based on preset time and space constraints, a density clustering algorithm is used to cluster the different nodes to obtain clustering results. Based on the clustering results, trajectory reconstruction is performed on the nodes within each cluster; wherein, the trajectory reconstruction algorithm connects the time points and spatial locations corresponding to the same unregistered person's images from different image acquisition points into a spatiotemporal trajectory. Multiple spatiotemporal trajectories of the same unregistered person are integrated into a unified trajectory database to construct a dynamically updated personnel trajectory map. The personnel trajectory map is matched and analyzed with the set sensitive areas to screen out the intersecting paths and analyze the abnormal behavior patterns of the trajectories, including dwell time, path complexity and area access frequency. For the identified abnormal behavior, the multimodal characteristics of the person, abnormal behavior conclusions and behavior analysis information are pushed to adjacent preset areas.

2. The method for personnel information security screening based on big data according to claim 1, characterized in that, A camera is set up in the first preset area to collect facial data images of people in the first preset area in real time. After the facial data images are preprocessed by image enhancement and noise reduction, a pre-trained facial recognition deep learning model is used to extract the corresponding feature vectors from the images. Each facial image is mapped to a feature vector of fixed length, and its vector value represents the feature of the corresponding face. The extracted feature vectors are then normalized.

3. The method for personnel information security screening based on big data according to claim 2, characterized in that, The preset personnel information database is a feature vector database extracted from pre-registered personnel. Based on the feature comparison algorithm, the similarity between the feature vector of the face data collected in the first preset area and the feature vector set in the preset personnel information database is calculated. According to the set similarity threshold, the corresponding face data with a similarity lower than the threshold is determined as the unregistered personnel.

4. The method for personnel information security screening based on big data according to claim 1, characterized in that, The target tracking method is used to analyze the real-time behavior trajectory of non-registered personnel within the first preset area, and the facial data features of non-registered personnel are used as identification features. By using multi-view mapping of cameras, the location of unregistered personnel is converted into two-dimensional plane coordinates. Combined with camera timestamps, the time points of each tracking record of the location of unregistered personnel are stored in time series form. The identification features of unregistered personnel are used to match trajectory points under multiple camera views and stitch together a complete trajectory. Based on temporal and spatial similarity, cluster analysis is performed on the trajectory points of unregistered personnel to identify whether loitering behavior exists.

5. The method for personnel information security screening based on big data according to claim 1, characterized in that, Within the second preset area, the appearance and behavioral features of non-registered personnel are extracted using target detection and segmentation technology, and the facial features, appearance features, and behavioral features of registered personnel are fused together based on an attention mechanism to generate the multimodal features; Based on multimodal features, images of the same unregistered person captured by different cameras are matched to form cross-regional associations. Each camera captures an image of an unregistered person as a node, and the node's attributes include timestamp, spatial location, and multimodal features. Set matching constraints by combining timestamps and camera spatial layout information; A similarity matrix is ​​constructed for all the nodes, and global matching is performed to associate multiple nodes of the same unregistered person.

6. The method for personnel information security screening based on big data according to claim 5, characterized in that, Based on the node matching results, the images of non-registered personnel are clustered to form cross-regional trajectories, specifically: The density threshold and minimum number of nodes are set based on the similarity between nodes and the established time and space constraints; Nodes that meet the density threshold and minimum number of nodes are clustered to form multiple clusters, each cluster representing the trajectory of the same unregistered person under multiple cameras.

7. The method for personnel information security screening based on big data according to claim 6, characterized in that, Based on the nodes contained in each cluster in the clustering results, they are sorted according to timestamps to form a time series; Between any two adjacent nodes, linear interpolation is performed based on time interval and spatial distance to generate refined trajectory points, which are then connected to form trajectory segments. Multiple trajectory segments within the same cluster are integrated, and then, for the integrated trajectory segments from multiple clusters of the same unregistered person, time series and adjacent camera position associations are used to stitch the trajectory segments into a complete spatiotemporal trajectory.

8. The method for personnel information security screening based on big data according to claim 1, characterized in that, Based on the sensitive areas set in the first and second preset areas, the constructed trajectory map of unregistered personnel is compared with the geographical location of the sensitive areas using cross-detection matching analysis based on the trajectory points and the boundaries of the sensitive areas. For detected trajectory paths that intersect with sensitive areas, the dwell time is extracted to determine whether the unregistered person has abnormal loitering behavior; By analyzing the path complexity, we can determine whether there are any abnormal detours or repetitive behaviors by unregistered personnel. By analyzing the frequency of access to their area, we can determine whether they exhibit abnormal behavior such as frequently entering and exiting sensitive areas.

9. A method for personnel information security screening based on big data according to claim 8, characterized in that, For the abnormal behaviors identified during screening, the behavioral analysis results, along with the multimodal characteristic information of the unregistered person, are pushed to an adjacent preset area in the form of a screening report; wherein, the screening report includes: Abnormal behavior conclusion, that is, whether the unregistered person has engaged in abnormal loitering, detours, or frequent entry into sensitive areas; Behavioral analysis information, namely the results of the analysis of the unregistered person's stay time and path complexity.

10. A personnel information security screening system based on big data, characterized in that, include: The face recognition module is configured as follows: Within a first preset area, facial data to be processed within the area is acquired. Based on image feature extraction, the facial data is compared with a dynamically updated preset personnel information database to filter out unregistered personnel. The behavior recognition module is configured as follows: For the unregistered personnel selected, the real-time behavior trajectory is tracked using the target tracking method, and their movement path and time sequence within the first preset area are recorded. Based on the preset time threshold and spatial conditions, it is determined whether the unregistered personnel have abnormal loitering behavior. The regional trajectory association module is configured as follows: Within a second preset area, multimodal features are extracted from multiple images of unregistered individuals. For images of the same unregistered individual from different image acquisition points, feature similarity is calculated for matching and association. Specifically, the features of the images of individuals acquired from different image acquisition points are used as nodes. Based on preset time and space constraints, a density clustering algorithm is employed to cluster these nodes, yielding clustering results. Based on these clustering results, trajectory reconstruction is performed on the nodes within each cluster. The trajectory reconstruction algorithm connects the time points and spatial locations corresponding to the same unregistered individual images from different image acquisition points to form a spatiotemporal trajectory. The abnormal behavior push module is configured as follows: Multiple spatiotemporal trajectories of the same unregistered person are integrated into a unified trajectory database to construct a dynamically updated personnel trajectory map. The personnel trajectory map is matched and analyzed with the set sensitive areas to screen out the intersecting paths and analyze the abnormal behavior patterns of the trajectories, including dwell time, path complexity and area access frequency. For the identified abnormal behavior, the multimodal characteristics of the person, abnormal behavior conclusions and behavior analysis information are pushed to adjacent preset areas.

Images