On-duty perimeter electronic sentry intelligent prevention and control system and method based on AI vision

Abstract

The invention relates to the technical field of intelligent security and protection monitoring, in particular to an intelligent prevention and control system and method for an on-duty perimeter electronic sentry based on AI vision, and the system comprises a full-dimensional sensing module which collects visible light, thermal imaging video streams and radar detection data of a perimeter area; the intelligent analysis module is used for carrying out behavior feature extraction on the video stream based on a deep learning model and generating a three-dimensional track vector of an intrusion target; the comprehensive management and control module maps the real-time track of the invasion target to a virtual electronic defense area for visual presentation; and the linkage processing module is used for scheduling the sound optical driver isolation equipment and the unmanned aerial vehicle countering unit according to the threat level logic resolving result. According to the invention, by fusing AI vision and a multi-dimensional perception technology, accurate identification and all-weather automatic tracking of perimeter intrusion behaviors are realized, the problems of dependence on manpower, existence of blind areas and response lag in traditional security and protection are solved, and the three-dimensional prevention and control level and emergency response speed of an on-duty area are greatly improved.

Description

AI Vision-Based Intelligent Perimeter Electronic Sentinel Control System and Method Technical Field

[0001] This invention relates to the field of intelligent security monitoring technology, and in particular to an intelligent perimeter electronic sentry system and method based on AI vision. Background Technology

[0002] Intelligent security monitoring technology refers to a technological system that utilizes computer vision, IoT sensing, big data analysis, and other methods to conduct all-weather monitoring, identification, and early warning of specific areas. It is widely used in perimeter protection of military bases, prisons, border defenses, and important facilities. This field primarily focuses on using technological means to replace or assist manual patrols, achieving the automatic detection and handling of security threats such as intrusion and sabotage. Traditional perimeter protection systems, on the other hand, rely mainly on passive defense systems consisting of physical walls, barbed wire, fiber optic cables, and fixed-point video surveillance cameras. They typically rely on manual review of video feeds at a monitoring center or simple physical triggering mechanisms, such as infrared beam blocking, to detect intrusion behavior.

[0003] However, existing technologies have significant drawbacks in practical operation. Relying solely on physical barriers and passive video recording prevents the system from actively identifying the specific behavioral attributes of targets. This makes it highly susceptible to adverse weather conditions, interference from small animals, or swaying trees, leading to numerous false alarms and causing staff to become complacent, as if crying wolf. Furthermore, traditional surveillance footage is disconnected from geographical space, making it difficult for command personnel to intuitively grasp the intruder's exact location and escape route. Moreover, various security subsystems, such as video, alarm, and broadcast systems, often operate independently, lacking deep logical linkage. This prevents the rapid deployment of comprehensive suppression measures such as sound and light attacks and drone tracking during emergencies, severely impacting the timeliness and accuracy of emergency response. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an AI vision-based intelligent perimeter electronic sentry control system and method. The technical solution is as follows: Firstly, an AI vision-based intelligent perimeter electronic sentry control system is provided. This system includes: a full-dimensional perception module configured to collect visible light video data, infrared thermal imaging data, and radar echo data of the monitored area in real time through a multispectral camera group and a millimeter-wave radar array deployed at the perimeter, and to perform timestamp synchronization processing on the data; and an intelligent analysis module configured to receive the data from the full-dimensional perception module, extract the skeletal key points and behavioral features of moving targets using a pre-set deep learning vision algorithm, and simultaneously analyze the radar echo data to calculate the target's distance, angle, and speed parameters. The radar target coordinates are mapped to the video pixel coordinate system through a coordinate system transformation matrix to generate fused target fingerprint information. The integrated control module is configured to load a three-dimensional geographic information model of the monitored area, map the fused target fingerprint information into the three-dimensional geographic information model, construct an AR panoramic real-scene map, and determine the intrusion status of the target based on preset virtual electronic perimeter rules, generating a graded early warning command. The linkage response module is configured to respond to the graded early warning command, automatically dispatch the corresponding sound and light strike equipment to perform strong light blinding or sound wave drive-away actions, and generate guidance commands to drive the PTZ camera to continuously track and shoot the locked target.

[0005] As a further aspect of the present invention, the intelligent analysis module performs the following logic when fusion of radar and video: constructing a homography transformation matrix between the radar coordinate system and the camera coordinate system, projecting the target spatial points detected by the radar onto the video image plane; calculating the overlap rate between the projected points and the target bounding boxes detected by video analysis; if the overlap rate is higher than a preset threshold, it is determined to be the same target, and the high-precision distance information of the radar and the high-precision category information of the video are attribute-associated; if the target is lost due to ambient light interference in video analysis, the camera pan-tilt unit is guided to rotate based on the Kalman filter prediction trajectory of the radar data until the visual target is recaptured.

[0006] As a further aspect of the present invention, the integrated control module includes virtual electronic perimeter delineation logic, specifically including: in the three-dimensional geographic information model, delineating multi-level virtual warning zones along the inner and outer sides of the physical perimeter wall, the virtual warning zones including early warning zones, warning zones, and response zones; when the coordinates of the fused target fingerprint information fall into different levels of the virtual warning zones, different levels of alarm strategies are triggered; the alarm strategies include: when the target enters the early warning zone, triggering a voice warning; when the target enters the warning zone, triggering an audible and visual flash and locking the target trajectory; when the target enters the response zone, triggering the blocking facilities and drone countermeasures process.

[0007] As a further embodiment of the present invention, the system also includes an unmanned aerial vehicle (UAV) air-to-ground coordination module, configured to: receive the GPS coordinates of the intrusion target sent by the integrated control module; schedule the UAV patrol unit to take off automatically and plan the optimal flight path to reach the target airspace; cut off the remote control link of the intrusion aircraft through the communication jamming equipment carried by the UAV, or conduct an aerial deterrent to the ground intruders through the airborne loudspeaker.

[0008] As a further embodiment of the present invention, the acoustic and optical strike equipment in the linkage response module includes an intelligent PTZ searchlight and a directional acoustic wave disperser; the linkage response module is configured to calculate the deflection angle and coverage range of the acoustic and optical strike equipment according to the target's moving speed and direction, and control the searchlight beam and acoustic wave beam to cover the target's current position in real time to implement non-contact blocking.

[0009] As a further embodiment of the present invention, the system also includes a duty terminal interaction module, configured to: push the AR panoramic real-scene map and alarm information to the sentry duty terminal and the mobile handheld terminal in real time; receive manual confirmation signals or one-click alarm signals from the terminals; when a one-click alarm signal is received, forcibly switch the monitoring videos of all related areas to the main control screen and broadcast the alarm information to all cooperating sentry posts.

[0010] As a further embodiment of the present invention, the system also includes a bullet displacement monitoring module, configured to: monitor the distance between the on-duty firearm and the displacement monitoring base station in real time through radio frequency identification technology; when the firearm leaves the set area or the connection signal is interrupted, immediately send a firearm off-duty alarm signal to the integrated control module, and link the nearest camera to capture the personnel holding the firearm.

[0011] As a further embodiment of the present invention, the system also includes a biometric verification module, configured to: verify the identity of personnel at the entrance and exit through facial recognition or fingerprint recognition technology; compare the verified personnel information with a local preset personnel list; and if an unauthorized person or an unknown person is found to be attempting to enter, trigger an illegal intrusion alarm and lock the access control.

[0012] As a further aspect of the present invention, the intelligent analysis module is also configured with abnormal behavior detection logic, specifically including: establishing a time series model of key points of the human skeleton, analyzing the posture changes of the duty personnel; when it is detected that the duty personnel are in a position with their eyes closed, head drooping, or lying down for a long time for more than a preset duration, it is determined to be an abnormality of sleeping on duty or leaving the post, and an alarm record of duty abnormality is generated.

[0013] On the other hand, the AI ​​vision-based intelligent perimeter electronic sentry control method, which is executed based on the aforementioned AI vision-based intelligent perimeter electronic sentry control system, includes the following steps: S1: synchronously collect perimeter environmental data through a multispectral camera and millimeter-wave radar; S2: extract target behavior features from the video using a deep learning algorithm and fuse radar spatial data to generate fused target fingerprint information; S3: load a three-dimensional geographic information model, map the fused target fingerprint information onto the three-dimensional geographic information model to construct an AR panoramic real-scene map, and determine its positional relationship with the virtual electronic perimeter; S4: when an illegal intrusion is determined, automatically dispatch sound and light attack equipment to physically drive away the target according to the intrusion area level, and guide the camera pan-tilt unit to continuously track the target.

[0014] The beneficial effects of the technical solution provided by the embodiments of the present invention include at least the following: The present invention, by constructing an electronic sentinel system based on AI vision and radar vision fusion, realizes the leap from "passive monitoring" to "active prevention and control" in perimeter security. By integrating the spatial perception capability of millimeter-wave radar with the visual recognition capability of cameras, it effectively filters false alarms caused by environmental factors. While ensuring all-weather coverage without blind spots, it achieves accurate positioning and continuous tracking of intrusion targets. Utilizing 3D AR panoramic map technology, it intuitively maps abstract alarm data onto a real-world map, greatly improving the intuitiveness of command and dispatch and decision-making efficiency. It establishes a deeply integrated sound and light strike and drone countermeasure mechanism, which can automatically implement non-contact physical blocking the moment a threat is detected, significantly shortening the response time from the discovery of an alarm to effective handling, and ensuring the absolute safety of the duty area. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 is a flowchart of the AI ​​vision-based intelligent perimeter electronic sentry control system provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] As shown in Figure 1, this embodiment of the invention provides an AI vision-based intelligent perimeter electronic sentry control system, which includes: a full-dimensional perception module configured to collect visible light video data, infrared thermal imaging data and radar echo data of the monitored area in real time through a multispectral camera group and a millimeter-wave radar array deployed at the perimeter, and to perform timestamp synchronization processing on the data.

[0020] The multispectral camera array comprises a visible light sensor and an uncooled infrared focal plane array detector, acquiring an RGB image stream with a resolution of 3840 x 2160 pixels and a thermal imaging data stream with a resolution of 640 x 512 pixels, respectively. Simultaneously, a 77 GHz millimeter-wave radar array transmits frequency-modulated continuous waves at a frequency of 20 frames per second and receives echo signals reflected from targets. To address the data asynchrony issue between heterogeneous sensors, the full-dimensional sensing module employs a hard synchronization mechanism based on a precise time protocol. Specifically, the master clock server broadcasts nanosecond-precision timing signals to the camera array and radar array via a fiber optic network. Each sensor writes the current system absolute timestamp into the data packet header when acquiring each frame. The data processing unit establishes a 500-millisecond sliding time window, sorting and matching the received video frames and radar point cloud data according to their timestamp values. When the timestamp of a video frame is T1, and there are two adjacent radar data frame timestamps Ta and Tb within the sliding window, satisfying the relationship Ta < T1 < Tb, linear interpolation is performed. For example, if Ta is 1000 milliseconds, Tb is 1050 milliseconds, and T1 is 1020 milliseconds, and the radar coordinates of the target at time Ta are (10, 20) and at time Tb are (12, 22), by calculating the time difference ratio, i.e., (1020 - 1000) divided by (1050 - 1000), an interpolation coefficient of 0.4 is obtained, and thus the synchronous radar coordinates corresponding to time T1 are calculated to be (10.8, 20.8). This strict alignment based on the time dimension eliminates the spatiotemporal deviation caused by different sensor sampling rates, ensuring the consistency of the data benchmark in subsequent fusion processing. Table 1 shows the configuration parameters and synchronization performance indicators of each sensor in the full-dimensional perception module.

[0021] Table 1 Sensor Configuration and Synchronization Parameters

[0022] As shown in Table 1, the full-dimensional sensing module provides a reliable data foundation for the system through high-precision hardware configuration and strict time synchronization protocol.

[0023] The intelligent analysis module is configured to receive data from the full-dimensional perception module, extract the skeletal key points and behavioral features of moving targets using a pre-built deep learning vision algorithm, and simultaneously analyze radar echo data to calculate the target's distance, angle, and velocity parameters. It then maps the radar target coordinates to the video pixel coordinate system using a coordinate system transformation matrix, generating fused target fingerprint information. When performing radar and video fusion, the following logic is executed: a homography transformation matrix is ​​constructed between the radar coordinate system and the camera coordinate system, projecting the target spatial points detected by the radar onto the video image plane; the overlap rate between the projected points and the target bounding boxes detected by video analysis is calculated; if the overlap is high... If the overlap ratio is higher than a preset threshold, it is determined to be the same target, and the high-precision distance information of the radar is associated with the high-precision category information of the video. If the target is lost due to interference from ambient light in the video analysis, the camera pan-tilt unit is guided to rotate based on the Kalman filter prediction trajectory of the radar data until the visual target is recaptured. It is also equipped with abnormal behavior detection logic, which includes: establishing a time series model of key points of the human skeleton to analyze the posture changes of the duty personnel; when it is detected that the duty personnel are in a position with their eyes closed, head drooping, or lying down for a long time for more than a preset duration, it is determined to be an abnormality of sleeping on duty or leaving the post, and an alarm record of duty abnormality is generated.

[0024] The intelligent analysis module preprocesses the input video stream, including denoising and size normalization, before feeding it into a convolutional neural network-based skeleton extraction model. This model comprises an input layer, a backbone feature extraction network, a cascaded stage module, and an output layer. The backbone network employs a residual network structure, extracting deep image features through consecutive 3x3 convolutional kernels and modified linear unit activation functions. The model's output layer generates heatmaps and affinity field vectors containing 18 human keypoints (such as nose, neck, shoulder, elbow, and wrist). For radar data, a fast Fourier transform is performed on the echo signal to calculate the target's range, azimuth, and radial velocity. In the fusion stage, a homography transformation matrix from the radar coordinate system to the image pixel coordinate system is obtained beforehand through calibration experiments. This matrix is ​​a 3x3 parametric matrix, calculated by selecting at least four control points with known correspondences in the scene. The target spatial coordinates (X, Y, Z) detected by the radar are constructed into homogeneous coordinate vectors and multiplied with the homography matrix to obtain the normalized image plane projection coordinates (u, v). Subsequently, the intersection-union ratio (IUGR) between the virtual bounding box constructed at the projection point and the target bounding box detected by the visual algorithm is calculated. A preset threshold of 0.6 is set, and the calculation logic is as follows: the area of ​​the intersection region between the virtual bounding box and the visual bounding box is divided by the area of ​​their union region. For example, if the visual detection box area is 10,000 pixels, the virtual bounding box area is 10,000 pixels, and the overlap area is 7,000 pixels, then the union area is 13,000 pixels, and the IUGR is 7,000 divided by 13,000, which is approximately 0.538. In this example, 0.538 is less than the threshold of 0.6, indicating that the two are not the same target or there is a large error. If the overlap area is 8,000 pixels, the union is 12,000 pixels, and the IUGR is 0.66, which is greater than 0.6, then they are considered the same target. The 150-meter distance value measured by radar is then bound to the "personnel" category label of the visual recognition to generate a unique fused target fingerprint. For abnormal behavior detection, a long short-term memory network is used to analyze the skeletal point sequence. The network structure includes input gates, forget gates, and output gates to handle long-term dependencies in time-series data. It continuously monitors the vertical angle between the line connecting the head and neck key points of on-duty personnel, as well as the aspect ratio of the eye key points. If the angle between the head-connecting line and the vertical direction is greater than 60 degrees (indicating head tilting or downward tilting), and the eye aspect ratio is less than 0.2 (indicating closed eyes), and this state lasts for more than a preset duration of 300 seconds, it is considered an abnormal sleep deprivation. If the vertical height between the center points of the shoulders and hips is detected to be less than 30% of the ground reference height, and this lasts for more than 60 seconds, it is considered an abnormal lying down or leaving the post.

[0025] The integrated control module is configured to load a 3D geographic information model of the monitored area, map the fused target fingerprint information into the 3D geographic information model, construct an AR panoramic real-scene map, and determine the intrusion status of the target based on preset virtual electronic perimeter rules, generating tiered early warning commands. It includes virtual electronic perimeter delineation logic, specifically: in the 3D geographic information model, multi-level virtual alert zones are delineated along the inner and outer sides of the physical perimeter wall, including warning zones, alert zones, and response zones; when the coordinates of the fused target fingerprint information fall into different levels of virtual alert zones, different levels of alarm strategies are triggered; alarm strategies include: when the target enters the warning zone, a voice warning is triggered; when the target enters the alert zone, an audible and visual flash is triggered and the target trajectory is locked; when the target enters the response zone, the blocking facilities and drone countermeasures are triggered.

[0026] The integrated control module first retrieves high-precision 3D model files of the monitored area from the geographic information system database. These files, in OBJ or FBX format, contain terrain textures and building geometry. Using the graphics processing unit's rendering pipeline, the 3D model is texture-mapped and perspective-overlaid with the real-time video stream, resulting in an augmented reality panoramic map on the screen. Within this map coordinate system, a raycasting algorithm is applied to determine the area of ​​the electronic fence. Specifically, the virtual alert zone is defined as a set of polygon vertices. A ray is drawn from the target's current coordinates in any direction, and the number of intersections between this ray and each side of the polygon is calculated. If the number of intersections is odd, the target is considered to be within the zone; if it is even, it is outside the zone. The hierarchy of the virtual alert zone is based on its distance from the physical fence: the warning zone is set to 50 meters to 20 meters outside the fence; the alert zone is set to 20 meters to 0 meters outside the fence and 0 meters to 10 meters inside; and the response zone is the core area within 10 meters inside the fence. For example, when the coordinates of the target's fingerprint display are 35 meters outside the perimeter wall, the ray-mapping algorithm determines that it is in the warning zone, immediately generates a Level 1 warning command, and calls the on-site broadcast to play the voice message "You have entered the monitored area, please leave immediately." As the target continues to approach, and the coordinates are updated to 10 meters outside the perimeter wall, it is determined that it has entered the restricted area, generating a Level 2 warning command, triggering strobe lights, and driving the pan-tilt unit to lock onto the target. If the target climbs over the perimeter wall and enters within 5 meters inside, it is determined that it has entered the disposal zone, generating a Level 3 highest command, and initiating the physical blocking and countermeasure procedures. Through hierarchical control, the false alarm rate is effectively reduced and a tiered response and linkage module for threats is configured to respond to hierarchical early warning commands, automatically dispatch corresponding acoustic and visual strike equipment to perform strong light blinding or sound wave repulsion actions, and generate guidance commands to drive the PTZ camera to continuously track and film the locked target; the acoustic and visual strike equipment includes intelligent PTZ searchlights and directional sound wave dispersers; the configuration is to calculate the deflection angle and coverage area of ​​the acoustic and visual strike equipment according to the target's moving speed and direction, and control the searchlight beam and sound wave beam to cover the target's current position in real time to implement non-contact blocking.

[0027] The coordinated response module primarily controls a high-intensity searchlight with a luminous flux of 20,000 lumens, and a directional acoustic wave disperser with a maximum sound pressure level of 150 dB. To achieve precision strikes, a built-in ballistic prediction algorithm (a simplified version for light / sound speed projection) is implemented. The target's current position coordinates (x, y) and velocity vector (vx, vy) are obtained, and combined with the device's response delay time t_delay (typically 0.5 seconds) and rotational angular velocity, the target's future predicted position (x_pred, y_pred) is calculated. The calculation logic is as follows: the predicted x-coordinate equals the current x-coordinate plus the product of the horizontal velocity and the delay time; the predicted y-coordinate is calculated similarly. Subsequently, the horizontal deflection angle and pitch angle from the device's current position (x0, y0) to the predicted position are calculated. The tangent of the horizontal deflection angle equals (y_pred minus y0) divided by (x_pred minus x0), and the angle value is obtained using the arctangent function. For example, if a target is located 100 meters due north of the device and moves eastward at a speed of 5 meters per second, with a device delay of 0.5 seconds, the predicted target position is 2.5 meters eastward. Calculating the arctangent (2.5 divided by 100) yields a required eastward deflection of approximately 1.43 degrees. This calculated angle is converted into a pulse-width modulation signal, driving the servo controller of the gimbal motor to precisely align the searchlight beam center with the sound wave beam center at the predicted target position. Simultaneously, the sound wave emission power is dynamically adjusted according to the target distance; the closer the distance, the lower the power output, to prevent permanent hearing damage and comply with the non-lethal rejection principle. Table 2 lists the audio-visual linkage response strategies at different distances.

[0028] Table 2 Joint Response Strategy Table

[0029] As shown in Table 2, based on the target distance parameters calculated in real time, the corresponding equipment working mode is automatically matched by looking up the table, thus realizing the automation and precision of the handling methods.

[0030] The UAV air-ground coordination module is configured to: receive the GPS coordinates of the intruding target sent by the integrated control module; schedule the UAV patrol unit to take off automatically and plan the optimal flight path to reach the target airspace; cut off the remote control link of the intruding aircraft through the communication jamming equipment carried by the UAV, or use the airborne loudspeaker to deter the ground intruders from the air.

[0031] When the integrated control module confirms a high-risk intrusion and ground equipment cannot provide complete coverage, the UAV air-ground coordination module will be triggered. This module first analyzes the latitude and longitude coordinates of the intruding target and, combined with current wind speed, terrain obstacle height, and no-fly zone data, uses the A-Star path planning algorithm to generate a flight trajectory. The A-Star algorithm calculates the sum of the actual cost from the starting point to the current node and the estimated cost from the current node to the destination, and searches for the node sequence with the minimum cost value in a rasterized map, thus avoiding buildings and trees. After takeoff, the UAV uses its onboard RTK positioning system to fly along the planned path, with positioning errors controlled within 10 centimeters. If the intruding target is a low-altitude aircraft (such as a black-flying UAV), the onboard full-band interference source is activated, emitting radio frequency noise signals covering the 2.4 GHz and 5.8 GHz bands, with a signal-to-noise ratio set to greater than 20 dB, forcing the intruding UAV to trigger either a loss-of-control return or a landing on the spot. If the target of the intrusion is ground personnel, the drone hovers 30 meters above the target, plays a deterrent message in a loop through its onboard loudspeaker, and transmits a video stream from the aerial perspective in real time.

[0032] The duty terminal interaction module is configured to: push AR panoramic real-scene maps and alarm information to the duty terminal and mobile handheld terminal in real time; receive manual confirmation signals or one-click alarm signals from the terminal; when a one-click alarm signal is received, force the monitoring video of all related areas to be switched to the main control screen and broadcast the alarm information to all cooperating guard posts.

[0033] The duty terminal interaction module uses the WebSocket full-duplex communication protocol to establish a persistent connection between the server and each terminal. AR map data streams, after H.265 encoding and compression, are pushed to the handheld terminal screens of Android or iOS systems at a rate of 25 frames per second. When an alarm is triggered, a JSON format data packet containing the alarm type, time, location, and a snapshot of the scene is constructed and pushed to the terminal notification bar. Once the duty personnel click the "One-Click Alarm" physical button or the virtual button on the screen, the terminal immediately sends a high-priority UDP data packet to the central server. Upon receiving this signal, the server-side interaction module immediately executes the video matrix switching command, traversing all online large-screen display controllers, forcibly retrieving the camera footage associated with the alarm area to the main view window, and highlighting it with a red border. Simultaneously, using text-to-speech technology, the alarm text information is converted into an audio stream, which is simultaneously played through the network broadcast system in the speakers of all coordinating sentry posts, ensuring that all sentries across the network are simultaneously informed of the alarm within one second.

[0034] The bullet displacement monitoring module is configured to: monitor the distance between the on-duty firearm and the displacement monitoring base station in real time through radio frequency identification technology; when the firearm leaves the set area or the connection signal is interrupted, it immediately sends a firearm off-duty alarm signal to the integrated control module and links the nearest camera to capture the image of the person holding the firearm.

[0035] Active RFID tags are embedded inside the stock of the firearms used on duty, and RFID reader / writer base stations are deployed in specific areas of the sentry post (such as gun cabinets or guard posts). The base station continuously reads the received signal strength indication value of the tag at 200-millisecond intervals. The received signal strength indication value is converted into physical distance using a logarithmic distance path loss model. The calculation logic is: the current received signal strength indication value equals the received signal strength indication value at a reference distance (1 meter) minus 10 times the path loss exponent multiplied by the logarithm of the distance. The distance is calculated in reverse. If the calculated distance value is greater than a set threshold (e.g., 5 meters), or if the base station fails to scan a specific tag ID for 5 consecutive reading cycles (i.e., within 1 second), it is determined that the firearm has been illegally moved. An alarm is immediately triggered, and the timestamp of the event and the base station ID are sent to the integrated control module. Based on the base station ID, the nearest PTZ camera is indexed, and the preset position is automatically adjusted to capture and record video of the area.

[0036] The biometric verification module is configured to: verify the identity of personnel at the entrance and exit through facial recognition or fingerprint recognition technology; compare the verified personnel information with the local pre-set personnel list; if an unauthorized person or an unknown person is found to be attempting to enter, trigger an illegal intrusion alarm and lock the access control.

[0037] The biometric verification module integrates a near-infrared binocular camera and a capacitive fingerprint scanner at the access control gate. When a person enters the recognition area, the camera captures a facial image and extracts a 128-dimensional facial feature vector using a lightweight convolutional neural network. The cosine similarity of this vector with a pre-stored database of feature vectors of on-duty personnel is calculated. The calculation logic is as follows: divide the dot product of the two vectors by the product of their magnitudes. If the calculated maximum similarity value is less than the system-set threshold of 0.85, or the fingerprint feature matching score is below 60, the person is identified as an unknown individual. At this time, a low-level signal is output to the access controller, forcing the electromagnetic lock to close, cutting off access permission, and simultaneously sending an unauthorized intrusion alarm event containing a captured photo to the backend.

[0038] The AI-vision-based intelligent perimeter electronic sentry control method is implemented based on the aforementioned AI-vision-based intelligent perimeter electronic sentry control system, and includes the following steps: S1: Simultaneously collect perimeter environmental data through multispectral cameras and millimeter-wave radar; S2: Extract target behavior features from the video using deep learning algorithms and fuse radar spatial data to generate fused target fingerprint information; S3: Load a 3D geographic information model, map the fused target fingerprint information onto the 3D geographic information model to construct an AR panoramic real-scene map, and determine its positional relationship with the virtual electronic perimeter; S4: When an illegal intrusion is detected, automatically dispatch sound and light attack equipment to physically drive away the target according to the intrusion area level, and guide the camera pan-tilt unit to continuously track the target.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI vision-based intelligent perimeter electronic sentry system for epidemic prevention and control, characterized in that: The system includes: a full-dimensional perception module, configured to collect visible light video data, infrared thermal imaging data, and radar echo data of the monitored area in real time through a multispectral camera group and millimeter-wave radar array deployed at the perimeter, and to perform time-stamp synchronization processing on the data; an intelligent analysis module, configured to receive the data from the full-dimensional perception module, extract the skeletal key points and behavioral features of the moving target using a preset deep learning vision algorithm, and simultaneously analyze the radar echo data to calculate the target's distance, angle, and velocity parameters, and map the radar target coordinates to the video pixel coordinate system through a coordinate system transformation matrix to generate fused target fingerprint information; a comprehensive management and control module, configured to load a three-dimensional geographic information model of the monitored area, map the fused target fingerprint information into the three-dimensional geographic information model, construct an AR panoramic real-scene map, and determine the intrusion status of the target based on preset virtual electronic perimeter rules, generating graded early warning commands; and a coordinated response module, configured to respond to the graded early warning commands, automatically dispatch corresponding sound and light strike equipment to perform strong light blinding or sound wave repulsion actions, and generate guidance commands to drive the PTZ camera to continuously track and film the locked target.

2. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, When performing radar and video fusion, the intelligent analysis module executes the following logic: It constructs a homography transformation matrix between the radar coordinate system and the camera coordinate system, projecting the target spatial points detected by the radar onto the video image plane; it calculates the overlap rate between the projected points and the target bounding boxes detected by the video analysis; if the overlap rate is higher than a preset threshold, it determines that they are the same target, and it associates the high-precision distance information of the radar with the high-precision category information of the video; if the video analysis loses the target due to ambient light interference, it guides the camera pan-tilt unit to rotate based on the Kalman filter prediction trajectory of the radar data until the visual target is recaptured.

3. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The integrated control module includes virtual electronic perimeter delineation logic, specifically including: in the three-dimensional geographic information model, multi-level virtual warning zones are delineated along the inner and outer sides of the physical perimeter wall, including a warning zone, a warning zone, and a response zone; when the coordinates of the fused target fingerprint information fall into different levels of the virtual warning zones, different levels of alarm strategies are triggered; the alarm strategies include: when the target enters the warning zone, a voice warning is triggered; when the target enters the warning zone, an audible and visual flash is triggered and the target trajectory is locked; when the target enters the response zone, the blocking facilities and drone countermeasures are triggered.

4. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The system also includes a UAV air-ground coordination module, configured to: receive the GPS coordinates of the intrusion target sent by the integrated control module; schedule the UAV patrol unit to take off automatically and plan the optimal flight path to reach the target airspace; cut off the remote control link of the intrusion aircraft through the communication jamming equipment carried by the UAV, or conduct an aerial deterrent to the ground intruders through the airborne loudspeaker.

5. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The sound and light attack equipment in the linkage response module includes an intelligent PTZ searchlight and a directional acoustic wave disperser; the linkage response module is configured to calculate the deflection angle and coverage range of the sound and light attack equipment according to the target's moving speed and direction, and control the searchlight beam and acoustic wave beam to cover the target's current position in real time to implement non-contact blocking.

6. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The system also includes a duty terminal interaction module, configured to: push the AR panoramic real-scene map and alarm information to the sentry duty terminal and mobile handheld terminal in real time; receive manual confirmation signals or one-click alarm signals from the terminal; when a one-click alarm signal is received, forcibly switch the monitoring video of all related areas to the main control screen and broadcast the alarm information to all cooperating sentry posts.

7. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The system also includes a bullet displacement monitoring module, configured to: monitor the distance between the on-duty firearm and the displacement monitoring base station in real time using radio frequency identification technology; when the firearm leaves the set area or the connection signal is interrupted, immediately send a firearm off-duty alarm signal to the integrated control module, and link the nearest camera to capture the image of the person holding the firearm.

8. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The system also includes a biometric verification module, configured to: verify the identity of personnel at the entrance and exit through facial recognition or fingerprint recognition technology; compare the verified personnel information with a local pre-set personnel list; if an unauthorized person or an unknown person is found to be attempting to enter, trigger an illegal intrusion alarm and lock the access control.

9. The AI ​​vision-based intelligent perimeter electronic sentry system for security and control as described in claim 1, characterized in that, The intelligent analysis module is also equipped with abnormal behavior detection logic, which specifically includes: establishing a time series model of key points of the human skeleton and analyzing the posture changes of the duty personnel; when it is detected that the duty personnel are in a position with their eyes closed, head drooping, or lying down for a long time for more than a preset time, it is determined to be an abnormality of sleeping on duty or leaving the post, and an alarm record of duty abnormality is generated.

10. A method for intelligent control of perimeter electronic sentry posts based on AI vision, characterized in that, The execution of the AI ​​vision-based intelligent perimeter electronic sentry system according to any one of claims 1-9 includes the following steps: S1: synchronously collecting perimeter environmental data through a multispectral camera and millimeter-wave radar; S2: extracting target behavior features from the video using a deep learning algorithm and fusing radar spatial data to generate fused target fingerprint information; S3: loading a three-dimensional geographic information model, mapping the fused target fingerprint information onto the three-dimensional geographic information model to construct an AR panoramic real-scene map, and determining its positional relationship with the virtual electronic perimeter; S4: when an illegal intrusion is determined, automatically dispatching sound and light attack equipment to physically drive away the target according to the intrusion area level, and guiding the camera pan-tilt unit to continuously track the target.

Images