Method and system for identifying intrusion in restricted area of power converter station

Abstract

The application provides a power converter station restricted area intrusion identification method and system, the method comprising: acquiring spatial structure information; according to a single camera tracking algorithm, performing track association on the same suspicious person in continuous frames to generate single camera person tracks in the field of view of each camera; according to the single camera person tracks and the spatial structure information, judging whether a boundary crossing event occurs for the suspicious person, if a boundary crossing event occurs, calculating a boundary crossing score of the boundary crossing event; calculating the appearance feature similarity of two single camera person tracks in the candidate cross-camera track pair, and calculating a cross-camera association score according to the appearance feature similarity; calculating a comprehensive intrusion score of the corresponding suspicious person, and when the comprehensive intrusion score exceeds a preset threshold, determining that the suspicious person constitutes a restricted area intrusion event. The application can reduce the false judgment rate of personnel intrusion in the restricted area of the power converter station.

Description

Technical Field

[0001] This invention relates to the field of personnel intrusion identification technology, and in particular to a method and system for intrusion identification in restricted areas of power converter stations. Background Technology

[0002] As the core hub of a high-voltage direct current (HVDC) transmission system, power converter stations contain high-voltage, high-current critical equipment in their confined areas (such as valve halls, control and protection rooms, and converter transformer areas), requiring extremely high levels of security for personnel access control. Any unauthorized entry into these areas could lead to equipment malfunctions, power outages, or even serious personal injury or death. Therefore, accurate and real-time intrusion detection in these confined areas is crucial for ensuring the safe and stable operation of converter stations.

[0003] In conventional security scenarios, facial recognition systems typically identify unauthorized personnel as intruders and trigger alarms upon detection within a monitored area. However, in the real-world monitoring environment of a power converter station's restricted area, even if the system identifies a person as a "suspicious individual" (i.e., an unauthorized person) through facial recognition, their presence does not necessarily equate to intrusion. The reasons are as follows: First, there are numerous public passageways and transitional areas surrounding the restricted area, meaning that the "appearance" of personnel does not equate to "entry." Surveillance cameras in the restricted area of ​​the converter station not only cover the area itself but also often cover its surrounding passageways, connecting corridors, equipment maintenance platforms, and other transitional spaces. While these spaces are within the monitoring field of view, they may not be within the boundaries of the restricted area. An unsuspecting individual might pass through a passageway outside the restricted area for normal plant operations or mistakenly enter a transitional area and immediately turn back, without actually crossing the restricted area boundary to enter the interior. If the system directly determines an intrusion based solely on "no suspicious personnel detected in the footage," it is clearly inconsistent with the facts and will generate numerous false alarms.

[0004] Second, the restricted area has long boundaries and numerous passageways, meaning personnel may move near the boundary without actually crossing it. The physical boundaries of the converter station's restricted area are complex, with fences, walls, doors, and windows often surrounding equipment inspection passages, pipeline corridors, and other spaces for personnel movement. Maintenance personnel may need to perform equipment checks, instrument readings, or temporary work outside the boundary. While these actions may make them appear close to the restricted area on the monitoring screen, they do not actually cross the boundary. Traditional solutions cannot distinguish the essential difference between "movements near the boundary" and "actually crossing the boundary."

[0005] Third, the restricted area contains multiple nested sub-areas of varying risk levels; an intrusion should not be considered complete without actual crossing. The restricted area of ​​a converter station is not a single, uniform space, but rather contains multiple nested sub-areas of different risk levels, such as the peripheral equipment area, the intermediate isolation area, and the core valve hall area, each with clearly defined physical or electronic boundaries. A non-suspicious individual might only be present in the outermost equipment platform area (which, while monitored, is not a strictly controlled core area), without crossing into a higher-risk inner area. Indiscriminately classifying all individuals within the restricted area as intruders is neither reasonable nor in line with the refined requirements of actual security management.

[0006] In summary, due to the special constraints of the converter station restricted area, such as "the transition space outside the monitoring coverage", "the complex boundary and the existence of designated activity space around it", and "the nesting of multiple risk sub-areas within the area", the "suspicious person" label generated by the failure of face comparison is only a trigger condition for initiating further observation, rather than a sufficient basis for determining intrusion, which leads to a high probability of misjudgment. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for identifying intrusions in restricted areas of power converter stations, aiming to solve at least one of the problems in the background art.

[0008] In a first aspect, the present invention provides a method for identifying intrusions in a restricted area of ​​a power converter station, the method comprising: Acquire multiple video streams deployed in the outer perimeter, boundary, and internal channels of the restricted area of ​​the power converter station, and obtain spatial structure information of the restricted area of ​​the power converter station; Target detection is performed on the multiple video streams to identify personnel targets in each frame of the multiple video streams. The detected personnel targets are compared with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. The trajectory of the same suspicious person in consecutive frames is associated according to the single camera tracking algorithm to generate the single camera personnel trajectory within the field of view of each camera. Based on the trajectory of the person in the single camera and the spatial structure information, it is determined whether the suspicious person has committed a boundary crossing event. If a boundary crossing event has occurred, the boundary crossing score of the boundary crossing event is calculated. Cross-camera matching is performed on the single-camera personnel trajectories of different cameras, and the single-camera personnel trajectories of different individuals matched with the same suspicious person are paired. Candidate cross-camera trajectory pairs are obtained based on the pairing results, and the appearance feature similarity of the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs is calculated. The cross-camera association score is calculated based on the appearance feature similarity. The comprehensive intrusion score of the corresponding suspicious person is calculated based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, the suspicious person is determined to constitute a restricted area intrusion event.

[0009] Secondly, the present invention provides an intrusion identification system for a restricted area of ​​a power converter station, the system comprising: The data acquisition module is used to acquire multiple video streams deployed in the outer perimeter, boundary and internal channels of the restricted area of ​​the power converter station, and to acquire the spatial structure information of the restricted area of ​​the power converter station; The target detection module is used to perform target detection on the multi-channel video streams to identify personnel targets in each frame of the multi-channel video streams. The detected personnel targets are compared with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. The trajectory of the same suspicious person in consecutive frames is associated according to the single-camera tracking algorithm to generate the single-camera personnel trajectory within the field of view of each camera. The boundary crossing detection module is used to determine whether a suspicious person has committed a boundary crossing event based on the trajectory of the person in the single camera and the spatial structure information. If a boundary crossing event occurs, the boundary crossing score of the boundary crossing event is calculated. The trajectory matching module is used to perform cross-camera matching of single-camera personnel trajectories from different cameras, and to pair different single-camera personnel trajectories that match the same suspicious person. Based on the pairing results, candidate cross-camera trajectory pairs are obtained, and the appearance feature similarity between the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs is calculated. Based on the appearance feature similarity, a cross-camera association score is calculated. The intrusion detection module is used to calculate the comprehensive intrusion score of the corresponding suspicious person based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, the suspicious person is determined to constitute a restricted area intrusion event.

[0010] Thirdly, the present invention provides a storage medium that stores one or more programs that, when executed by a processor, implement the above-described method for identifying intrusions in restricted areas of power converter stations.

[0011] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the above-mentioned method for identifying intrusions in the restricted area of ​​a power converter station.

[0012] Compared with the prior art, the present invention has the following advantages: This invention effectively eliminates data interference from a large number of compliant personnel and reduces the system's computational burden by first screening authorized personnel through facial recognition and then analyzing only those without suspicious intent. Building on this, it stitches together discrete observation segments of the same suspicious person from multiple non-overlapping camera views into a complete activity trajectory through single-camera trajectory tracking and cross-camera trajectory association. This overcomes the problem of lost tracking in the restricted area of ​​the converter station due to non-overlapping camera views, enabling continuous tracking of the entire movement process of suspicious persons. By introducing a signed distance function and a hysteresis region judgment mechanism, it accurately identifies whether suspicious persons actually cross the boundary of the restricted area (rather than just moving near the boundary). This avoids the misjudgment that equates "appearing in the surveillance footage" with "intrusion" in traditional solutions. Furthermore, by comprehensively calculating boundary crossing scores, cross-camera correlation scores, key equipment proximity scores, and restricted area dwell scores, a multi-dimensional quantitative intrusion scoring system is constructed. This allows the system to make progressive judgments based on weak evidence from facial recognition, combined with behavioral trajectory evidence, effectively balancing the detection rate and false alarm rate. Finally, an intrusion alarm is triggered only when accumulated evidence exceeds a preset threshold, significantly reducing the high false alarm rate caused by special scenario constraints such as "surveillance coverage perimeter transition space," "complex boundaries with surrounding activity space," and "multi-layered risk sub-regions within the area." Attached Figure Description

[0013] Figure 1 This is a flowchart of a restricted area intrusion identification method for power converter stations according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a restricted area intrusion identification system for a power converter station according to an embodiment of the present invention.

[0014] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0016] like Figure 1 As shown, an embodiment of the present invention proposes a method for intrusion identification in a restricted area of ​​a power converter station. The method includes steps S101 to S105, wherein: Step S101: Obtain multiple video streams deployed in the outer perimeter, boundary, and internal channels of the restricted area of ​​the power converter station, and obtain the spatial structure information of the restricted area of ​​the power converter station; It should be noted that by deploying multiple surveillance cameras at key locations such as the periphery, boundaries, and internal passages of the restricted area, multi-view, multi-coverage video stream data covering the entire restricted area and surrounding transitional spaces is obtained. This ensures that any person appearing inside the restricted area or near its boundaries can be captured by at least one camera. At the same time, the spatial structure information of the restricted area is obtained, including structured data such as the location of the restricted area boundaries, the range of the buffer zone, the boundaries of the key equipment area, the spatial coverage of each camera, and the spatial adjacency relationship between cameras. This provides an accurate geospatial reference framework for subsequent spatial positioning of suspicious persons, boundary crossing detection, and cross-camera trajectory correlation, enabling visual detection results to be mapped to the actual physical space for analysis.

[0017] Step S102: Perform target detection on the multi-channel video stream to identify personnel targets in each frame of the multi-channel video stream. Compare the detected personnel targets with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. Based on the single-camera tracking algorithm, the trajectory of the same suspicious person in consecutive frames is associated to generate the single-camera personnel trajectory within the field of view of each camera. In this step, the spatial structure information includes at least the boundaries of the restricted area, the boundaries of the buffer area, the boundaries of the critical equipment area, the coverage area of ​​each camera, and the spatial adjacency relationship between the cameras.

[0018] Specifically, the process begins by performing preprocessing operations on multiple video streams, including video frame extraction, time synchronization, noise suppression, size standardization, and image stabilization, to eliminate time discrepancies and image quality differences between the multiple video streams. Then, a trained target detection model is used to detect people in each frame, outputting the bounding box, detection confidence score, and category information for each person target, while filtering out low-quality detection results with confidence scores below a preset threshold. Based on this, the detected people targets are compared with a pre-set face database, which includes face images of all individuals with access rights to the restricted area of ​​the power converter station. Individuals who fail the comparison are marked as suspicious.

[0019] It should be noted that the target detection model can adopt a target detection network based on convolutional neural networks. Its training process involves collecting multiple video frame images from the actual monitoring scenario of the restricted area of ​​the converter station as training samples, labeling the bounding boxes and category labels of personnel targets in each frame image, and training the initial detection network using the labeled training samples. The network parameters are iteratively updated through backpropagation of the loss function until the loss converges or the preset training rounds are reached, resulting in a trained target detection model. This model is deployed in the restricted area intrusion identification system of the power converter station, and performs real-time inference on each frame image in the processed multi-channel video stream, outputting the bounding boxes, detection confidence, and category information of personnel targets in each frame image, and filtering out personnel targets with detection confidence below the preset confidence threshold.

[0020] Furthermore, in some embodiments, it is also necessary to perform trajectory association on the same suspicious person in consecutive frame images detected by the same camera, generate single-camera person trajectories within the field of view of each camera, and calculate the trajectory reliability of each single-camera person trajectory: ; in, For trajectory reliability, This represents the total number of frames that successfully detected suspicious individuals in the trajectory of a single camera. This represents the total number of continuous frames for tracing the movement of a person using a single camera. This represents the maximum number of consecutive missed frames in a single camera's personnel trajectory. The threshold for the number of consecutive missed frames allowed.

[0021] The reliability of the trajectory is directly used as a weighting factor for the boundary crossing score. This ensures that the crossing behavior is only given a valid score when the trajectory data supporting the boundary crossing judgment has sufficient integrity and continuity. This effectively avoids false crossing misjudgments caused by poor tracking quality and discontinuous or incomplete trajectories, and significantly improves the accuracy and robustness of the intrusion identification method in the complex monitoring environment of the converter station.

[0022] Step S103: Determine whether the suspicious person has committed a boundary crossing event based on the single camera personnel trajectory and the spatial structure information. If a boundary crossing event has occurred, calculate the boundary crossing score of the boundary crossing event. It should be noted that in this step, the grounding position or the center of the bottom edge of the bounding box of the suspicious person in the single camera's personnel trajectory is first defined as the person's position x, and the signed distance from the person's position x to the boundary of the restricted area is defined according to the following formula: ; in, Let x be the signed distance from the personnel's location to the boundary of the restricted area. Restricted area For the boundary of the restricted area, For personnel location The shortest distance to the boundary of the restricted area; if personnel are within the restricted area. If the value is positive, it means the person is outside the restricted area. It is a negative value; A hysteresis region of width h is set on both the outer and inner sides of the restricted area boundary. If a suspicious person satisfies the following condition for m consecutive frames before crossing the restricted area... Then, for n consecutive frames, the following condition is met. If the overall movement direction between consecutive m frames and consecutive n frames points towards the interior of the restricted area, then the suspicious person is deemed to have committed a valid inward boundary crossing event, and the boundary crossing score is calculated according to the following formula: ; in, This is the boundary crossing event indicator value, which is used when an inward boundary crossing event occurs. ,otherwise , Scoring is given for crossing the boundary.

[0023] In summary, by using the grounding position or the center of the bottom edge of the bounding box of a suspicious person as their spatial location, and combining the precise coordinates of the boundary of the restricted area in the spatial structure information, the signed distance from this position to the boundary of the restricted area is calculated. A positive value indicates that the person is inside the boundary, and a negative value indicates that they are outside the boundary. To avoid false judgments due to boundary jitter caused by camera positioning noise, hysteresis zones of a certain width are set on both the inner and outer sides of the boundary. Only when the suspicious person is outside the hysteresis zone outside the boundary for multiple consecutive frames from the outside, and then inside the hysteresis zone inside the boundary for multiple consecutive frames, and the overall movement direction during this period points towards the inside of the restricted area, is a valid inward boundary crossing event determined. For the trajectory determined to be a valid crossing, the boundary crossing score is calculated based on the reliability of the trajectory, thus realizing a quantitative assessment of the reliability of the crossing behavior.

[0024] Step S104: Perform cross-camera matching on the single-camera personnel trajectories of different cameras, and pair the different single-camera personnel trajectories that match the same suspicious person, obtain candidate cross-camera trajectory pairs based on the pairing results, calculate the appearance feature similarity of the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs, and calculate the cross-camera association score based on the appearance feature similarity. It should be noted that this step, by stitching together local trajectory fragments within the field of view of a single camera into a cross-camera global trajectory link, overcomes the problem of fragmented movement caused by the non-overlapping field of view of cameras in the converter station. This enables the system to continuously identify suspicious persons as they move between different camera fields of view and obtain their complete spatial movement sequence, providing global behavioral data support for comprehensive intrusion determination.

[0025] Specifically, the spatial adjacency relationship between each camera is first obtained, and adjacent camera pairs are determined based on the spatial adjacency relationship. It is assumed that an adjacent camera pair includes cameras... and cameras Calculate the camera's performance using the following formula. Single camera personnel trajectory and cameras Single camera personnel trajectory migration time difference : ; in, Personnel trajectory for single camera The start time, Personnel trajectory for single camera The end time; Only when Falling from arrive When the directed topological path P corresponds to the preset migration time range, then... and Generate candidate cross-camera trajectory pairs and calculate migration reachability scores: ; in, for and The migration reachability score, , These represent the expected migration time and migration timescale of the directed topological path P, respectively. For trajectory The location and trajectory of disappearance The normalized offset of the location of the occurrence relative to the corresponding entrance / exit gate area; This is the path position deviation coefficient; Let P be the number of directed edges contained in path P; To take values ​​in The penalty coefficient for missing intermediate observations within the range; Trajectories Final stage and trajectory The direction of movement in the first segment; These represent the travel directions of the first and last segments of path P, respectively. From the trajectory and Extract facial recognition features from individuals separately and The cross-camera correlation score is calculated using the following formula: ; ; in, for and Cross-camera correlation score, for and The similarity of appearance features, where T is the transpose.

[0026] It should be noted that all facial recognition comparison processes involved in this embodiment are carried out under the premise of obtaining user authorization and complying with security management regulations: the preset facial database used by the system only includes facial images of personnel who have been officially authorized by the power converter station management and have access to restricted areas, and the collection, storage and comparison of relevant facial images have all obtained the informed consent of the personnel being collected in advance, and comply with the relevant regulations on information security and personal privacy protection in the power industry, and are limited to the security control purpose of restricted area intrusion identification, and are not used for other purposes.

[0027] In summary, by introducing migration time consistency verification and a refined migration reachability scoring mechanism between adjacent cameras, efficient screening and accurate association of cross-camera trajectory pairing are achieved. The core effects are reflected in the following three aspects: First, by limiting the formation of candidate trajectory pairs to only those whose migration time difference falls within a preset migration time range corresponding to the directed topological path between the two cameras, the original globally traversed cross-camera matching search space is significantly compressed into calculations only for spatiotemporally adjacent candidate pairs, significantly reducing the computational complexity of cross-camera association and meeting the real-time processing requirements of multiple video streams in the converter station; Second, the migration reachability score comprehensively considers the degree of migration time consistency and the deviation of the entry / exit position relative to the passageway entrance. Multi-dimensional factors such as displacement, intermediate observation missing penalty, and consistency between movement direction and channel orientation enable the system to effectively determine whether two trajectories truly constitute a reasonable physical migration in space and time, avoiding a large number of false associations caused by matching based solely on appearance similarity. Thirdly, the system finally obtains the cross-camera association score by geometrically averaging the migration reachability score and the facial appearance feature similarity extracted from the two trajectories, achieving dual verification of "spatiotemporal rationality" and "visual similarity". This ensures that the two trajectories are verified simultaneously in two dimensions: "physically feasible" and "identity-matching", significantly improving the accuracy and robustness of cross-camera personnel identity association and providing a highly reliable trajectory splicing basis for subsequent comprehensive intrusion judgment.

[0028] Step S105: Calculate the comprehensive intrusion score of the corresponding suspicious person based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, it is determined that the suspicious person constitutes a restricted area intrusion event.

[0029] In this step, the individual camera movements of the same suspicious person are first connected sequentially in chronological order to form a cross-camera activity chain. ,in, , , These are the 1st, 2nd, and Mth single-camera personnel trajectories in the cross-camera activity chain, respectively, where M is the number of single-camera personnel trajectories included in the cross-camera activity chain C. Then, the cross-camera continuity score of the cross-camera activity chain C is calculated according to the following formula. : ; in, The cross-camera association score between the (k-1)th and kth single-camera personnel trajectories in the cross-camera activity chain C; Next, the shortest distance between the positions of all valid personnel in the cross-camera activity chain C and the protected area of ​​the critical equipment is calculated, and the critical equipment proximity score is calculated based on the shortest distance: ; in, Personnel location x is in the critical equipment protection area. Internal time Take 0, For the device proximity distance threshold, For distance transition scale, The approach score for critical equipment across camera activity chain C is used; in some embodiments, a distance transition scale is used to control the steepness of the approach score for critical equipment as the shortest distance changes. Its value can be determined according to the safety level requirements of the protection area of ​​critical equipment in the converter station. The higher the safety level, the smaller the value, making the score more sensitive to changes in distance.

[0030] It should be noted that the "critical equipment protection area" refers to the three-dimensional spatial range within the restricted area of ​​a power converter station, surrounding the core critical equipment, and requiring strict protection against unauthorized access by personnel. Its boundaries can be pre-defined and stored in the spatial structure information based on the actual equipment layout and safety control requirements of the converter station, serving as the spatial reference for subsequent calculation of critical equipment proximity scores. Specifically, this area can be, for example, a closed area formed by extending a preset safety distance outward from the outer contour of core equipment such as converter valves, converter transformers, and control and protection cabinets. During calculation, the shortest distance between the positions of all valid personnel in the cross-camera activity chain and this critical equipment protection area is used as a metric. When a personnel position is within the protection area, the shortest distance is 0; when a personnel position is outside the protection area, the shortest distance is the shortest Euclidean distance from the personnel position to the boundary of the protection area. Specifically, the critical equipment protection area is a pre-defined closed spatial area formed by extending a preset safety distance outward from the outer contour of the core equipment of the converter station, and its boundaries are used for distance calculation with respect to personnel positions.

[0031] According to cross-camera activity chain The current observation period is defined by the appearance and termination times of the suspected individual in the restricted area. The time interval between the current observation period and all historical observation periods of the same day and the previous day is obtained based on the termination time. The duration of all historical observation periods with time intervals less than a preset threshold is added to the duration of the current observation period to obtain the cumulative stay duration. The restricted area stay score is then calculated using the following formula: ; in, Score for staying in the restricted area. To calculate the total length of stay, The time scale is the duration of stay; The overall intrusion score is calculated using the following formula: ; in, To calculate the overall intrusion score, , , , All are non-negative weighting coefficients.

[0032] By constructing a complete cross-camera activity chain and introducing a multi-dimensional weighted fusion mechanism, a comprehensive quantitative assessment of the intrusion risk of suspicious personnel is achieved. Specifically, the scheme first stitches together all single-camera trajectories belonging to the same suspicious person in chronological order to form a complete cross-camera activity chain, and calculates the continuity score of the entire activity chain (i.e., the geometric mean of the cross-camera association scores of each adjacent trajectory), thereby providing a holistic measure of the stitching confidence of the entire activity chain and ensuring that all subsequent analyses are based on reliable trajectory stitching. On this basis, the shortest distance between the locations of all personnel in the activity chain and the protection area of ​​critical equipment is further calculated and converted into a critical equipment proximity score, enabling the system to quantitatively assess the potential security threat posed by suspicious personnel to core equipment. At the same time, by accumulating the duration of the suspicious person's stay in the restricted area and calculating the stay score, the system effectively identifies two different threat levels of behavioral patterns: "brief intrusion followed by rapid departure" and "deliberate long-term stay." Ultimately, the four indicators—boundary crossing score, cross-camera continuity score, critical equipment proximity score, and area stay score—are non-negatively weighted and fused. This upgrades intrusion detection from a single-dimensional "whether the boundary was crossed" to a multi-dimensional risk assessment system that comprehensively considers "how to enter, whether to continue penetrating, whether to approach core equipment, and whether to stay for an extended period." An alarm is only triggered when the evidence corroborates each other from different angles and the overall score exceeds a threshold, significantly improving the accuracy and reliability of intrusion identification. At the same time, the four weight coefficients can be flexibly configured according to the actual security level requirements of different areas of the converter station, demonstrating good scenario adaptability.

[0033] In summary, this invention effectively eliminates data interference from a large number of compliant personnel and reduces the system's computational burden by first screening authorized personnel through facial recognition and then analyzing only those without suspicious intent. Building on this, it stitches together discrete observation segments of the same suspicious person from multiple non-overlapping camera views into a complete activity trajectory through single-camera trajectory tracking and cross-camera trajectory association. This overcomes the problem of lost tracking in the restricted area of ​​the converter station due to non-overlapping camera views, achieving continuous tracking of the entire movement process of suspicious persons. Furthermore, by introducing a signed distance function and a hysteresis region judgment mechanism, it accurately identifies whether suspicious persons actually cross the boundary of the restricted area (rather than just moving near the boundary). This avoids the misjudgment that equates "appearing in the surveillance footage" with "intrusion" in traditional solutions. Furthermore, by comprehensively calculating boundary crossing scores, cross-camera correlation scores, key equipment proximity scores, and restricted area dwell scores, a multi-dimensional quantitative intrusion scoring system is constructed. This allows the system to make progressive judgments based on weak evidence from facial recognition, combined with behavioral trajectory evidence, effectively balancing the detection rate and false alarm rate. Finally, an intrusion alarm is triggered only when accumulated evidence exceeds a preset threshold, significantly reducing the high false alarm rate caused by special scenario constraints such as "surveillance coverage of peripheral transitional spaces," "complex boundaries with surrounding activity spaces," and "multi-layered risk sub-regions within the area."

[0034] like Figure 2 As shown, one embodiment of the present invention proposes an intrusion identification system for a restricted area of ​​a power converter station, the system comprising: The data acquisition module 10 is used to acquire multiple video streams deployed in the outer perimeter, boundary and internal channels of the restricted area of ​​the power converter station, and to acquire the spatial structure information of the restricted area of ​​the power converter station. The target detection module 20 is used to perform target detection on the multi-channel video stream to identify personnel targets in each frame of the multi-channel video stream. The detected personnel targets are compared with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. The trajectory of the same suspicious person in consecutive frames is associated according to the single-camera tracking algorithm to generate the single-camera personnel trajectory within the field of view of each camera. The boundary crossing detection module 30 is used to determine whether a suspicious person has committed a boundary crossing event based on the trajectory of the person in the single camera and the spatial structure information. If a boundary crossing event occurs, the boundary crossing score of the boundary crossing event is calculated. The trajectory matching module 40 is used to perform cross-camera matching on single-camera personnel trajectories from different cameras, and to pair different single-camera personnel trajectories that match the same suspicious person, and to obtain candidate cross-camera trajectory pairs based on the pairing results, and to calculate the appearance feature similarity between the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs, and to calculate the cross-camera association score based on the appearance feature similarity. The intrusion identification module 50 is used to calculate the comprehensive intrusion score of the corresponding suspicious person based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, the suspicious person is determined to constitute a restricted area intrusion event.

[0035] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described method for identifying intrusions in restricted areas of power converter stations.

[0036] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the above-mentioned power converter station restricted area intrusion identification method.

[0037] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0038] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0039] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0040] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A method for identifying intrusions in restricted areas of power converter stations, characterized in that, The method includes: Acquire multiple video streams deployed in the outer perimeter, boundary, and internal channels of the restricted area of ​​the power converter station, and obtain spatial structure information of the restricted area of ​​the power converter station; Target detection is performed on the multiple video streams to identify personnel targets in each frame of the multiple video streams. The detected personnel targets are compared with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. The trajectory of the same suspicious person in consecutive frames is associated according to the single camera tracking algorithm to generate the single camera personnel trajectory within the field of view of each camera. Based on the trajectory of the person in the single camera and the spatial structure information, it is determined whether the suspicious person has committed a boundary crossing event. If a boundary crossing event has occurred, the boundary crossing score of the boundary crossing event is calculated. Cross-camera matching is performed on the single-camera personnel trajectories of different cameras, and the single-camera personnel trajectories of different individuals matched with the same suspicious person are paired. Candidate cross-camera trajectory pairs are obtained based on the pairing results, and the appearance feature similarity of the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs is calculated. The cross-camera association score is calculated based on the appearance feature similarity. The comprehensive intrusion score of the corresponding suspicious person is calculated based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, the suspicious person is determined to constitute a restricted area intrusion event.

2. The method for identifying intrusions in confined areas of power converter stations according to claim 1, characterized in that, The spatial structure information includes at least the boundaries of the restricted area, the boundaries of the buffer area, the boundaries of the critical equipment area, the coverage area of ​​each camera, and the spatial adjacency relationship between the cameras. The step of performing target detection on the multiple video streams to identify human targets in each frame of the multiple video streams includes: The multiple video streams are sequentially processed by video frame extraction, time synchronization, noise suppression, size standardization, invalid frame removal, and image stabilization to obtain the processed multiple video streams. The trained target detection model is used to detect each frame of the processed multi-video stream, and outputs the bounding box, detection confidence and category information of the person target in each frame, and filters out the person targets whose detection confidence is lower than the preset confidence threshold.

3. The method for identifying intrusions in restricted areas of power converter stations according to claim 2, characterized in that, The step of generating a single-camera person trajectory within the field of view of each camera by associating the trajectory of the same suspicious person in consecutive frames based on a single-camera tracking algorithm includes: Trajectory association is performed on the same suspicious person in consecutive frames detected by the same camera to generate single-camera person trajectories within the field of view of each camera, and the reliability of each single-camera person trajectory is calculated: ； in, For trajectory reliability, This represents the total number of frames that successfully detected suspicious individuals in the trajectory of a single camera. This represents the total number of continuous frames for tracing the movement of a person using a single camera. This represents the maximum number of consecutive missed frames in a single camera's personnel trajectory. The threshold for the number of consecutive missed frames allowed.

4. The method for identifying intrusions in restricted areas of power converter stations according to claim 3, characterized in that, The step of determining whether a suspicious person has committed a boundary crossing event based on the single camera's trajectory and the spatial structure information, and calculating the boundary crossing score if such an event occurs, includes: Define the grounding position or the center of the bottom edge of the bounding box of the suspicious person in the single camera's personnel trajectory as the person's position x, and define the signed distance from the person's position x to the boundary of the restricted area according to the following formula: ； in, Let x be the signed distance from the personnel's location to the boundary of the restricted area. Restricted area For the boundary of the restricted area, For personnel location The shortest distance to the boundary of the restricted area; if personnel are within the restricted area. If the value is positive, it means the person is outside the restricted area. It is a negative value; A hysteresis region of width h is set on both the outer and inner sides of the restricted area boundary. If a suspicious person satisfies the following condition for m consecutive frames before crossing the restricted area... Then, for n consecutive frames, the following condition is met. If the overall movement direction between consecutive m frames and consecutive n frames points towards the interior of the restricted area, then the suspicious person is deemed to have committed a valid inward boundary crossing event, and the boundary crossing score is calculated according to the following formula: ； in, This is the boundary crossing event indicator value, which is used when an inward boundary crossing event occurs. ,otherwise , Scoring is given for crossing the boundary.

5. The method for identifying intrusions in restricted areas of power converter stations according to claim 4, characterized in that, The steps of performing cross-camera matching of single-camera personnel trajectories from different cameras, pairing different single-camera personnel trajectories that match the same suspicious person, obtaining candidate cross-camera trajectory pairs based on the pairing results, calculating the appearance feature similarity between the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs, and calculating the cross-camera association score based on the appearance feature similarity include: Obtain the spatial adjacency relationship between each camera, and determine adjacent camera pairs based on the spatial adjacency relationship. Assume that an adjacent camera pair includes cameras. and cameras Calculate the camera's performance using the following formula. Single camera personnel trajectory and cameras Single camera personnel trajectory migration time difference : ； in, Personnel trajectory for single camera The start time, Personnel trajectory for single camera The end time; Only when Falling from arrive When the directed topological path P corresponds to the preset migration time range, then... and Generate candidate cross-camera trajectory pairs and calculate migration reachability scores: ； in, for and The migration reachability score, , These represent the expected migration time and migration timescale of the directed topological path P, respectively. For trajectory The location and trajectory of disappearance The normalized offset of the location of the occurrence relative to the corresponding entrance / exit gate area; This is the path position deviation coefficient; Let P be the number of directed edges contained in path P; To take values ​​in The penalty coefficient for missing intermediate observations within the range; Trajectories Final stage and trajectory The direction of movement in the first segment; These represent the travel directions of the first and last segments of path P, respectively. From the trajectory and Extract facial recognition features from individuals separately and The cross-camera correlation score is calculated using the following formula: ； ； in, for and Cross-camera correlation score, for and The similarity of appearance features, where T is the transpose.

6. The method for identifying intrusions in restricted areas of power converter stations according to claim 5, characterized in that, The step of calculating the comprehensive intrusion score of the corresponding suspicious person based on the boundary crossing score and the cross-camera association score, and determining that the suspicious person constitutes a restricted area intrusion event when the comprehensive intrusion score exceeds a preset threshold, includes: Connect the individual camera movements of the same suspicious person in chronological order to form a cross-camera activity chain. ,in, , , These are the 1st, 2nd, and Mth single-camera personnel trajectories in the cross-camera activity chain, respectively, where M is the number of single-camera personnel trajectories included in the cross-camera activity chain C. The cross-camera continuity score of the cross-camera activity chain C is calculated using the following formula. : ； in, The cross-camera association score between the (k-1)th and kth single-camera personnel trajectories in the cross-camera activity chain C; Calculate the shortest distance between the locations of all valid personnel in the cross-camera activity chain C and the protected area of ​​critical equipment, and calculate the critical equipment proximity score based on the shortest distance: ； in, Personnel location x is in the critical equipment protection area. Internal time Take 0, For the device proximity distance threshold, For distance transition scale, For the key equipment proximity score in the cross-camera activity chain C; Based on the cross-camera activity chain C, the appearance time and termination time of the corresponding suspicious person in the restricted area constitute the current observation period. The time interval between the current observation period and all historical observation periods within the same day and the previous day is obtained based on the termination time. The duration of all historical observation periods with time intervals less than a preset threshold is added to the duration of the current observation period to obtain the cumulative stay time. The restricted area stay score is then calculated using the following formula: ； in, Score for staying in the restricted area. To calculate the total length of stay, The time scale is the duration of stay; The overall intrusion score is calculated using the following formula: ； in, To calculate the overall intrusion score, , , , All are non-negative weighting coefficients.

7. A restricted area intrusion identification system for a power converter station, characterized in that, The system includes: The data acquisition module is used to acquire multiple video streams deployed in the outer perimeter, boundary and internal channels of the restricted area of ​​the power converter station, and to acquire the spatial structure information of the restricted area of ​​the power converter station; The target detection module is used to perform target detection on the multi-channel video streams to identify personnel targets in each frame of the multi-channel video streams. The detected personnel targets are compared with a preset face database, which includes face images of all persons with access rights to the restricted area of ​​the power converter station. If the comparison fails, the personnel target is determined to be a suspicious person. The trajectory of the same suspicious person in consecutive frames is associated according to the single-camera tracking algorithm to generate the single-camera personnel trajectory within the field of view of each camera. The boundary crossing detection module is used to determine whether a suspicious person has committed a boundary crossing event based on the trajectory of the person in the single camera and the spatial structure information. If a boundary crossing event occurs, the boundary crossing score of the boundary crossing event is calculated. The trajectory matching module is used to perform cross-camera matching of single-camera personnel trajectories from different cameras, and to pair different single-camera personnel trajectories that match the same suspicious person. Based on the pairing results, candidate cross-camera trajectory pairs are obtained, and the appearance feature similarity between the two single-camera personnel trajectories in the candidate cross-camera trajectory pairs is calculated. Based on the appearance feature similarity, a cross-camera association score is calculated. The intrusion detection module is used to calculate the comprehensive intrusion score of the corresponding suspicious person based on the boundary crossing score and the cross-camera association score. When the comprehensive intrusion score exceeds a preset threshold, the suspicious person is determined to constitute a restricted area intrusion event.

8. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the power converter station restricted area intrusion identification method as described in any one of claims 1-6.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the power converter station restricted area intrusion identification method according to any one of claims 1-6.

Images