Intelligent remote control method and system for unmanned intelligent turntable

Abstract

The application provides an intelligent remote control method and system for an unmanned intelligent rotating table. The method discloses an intelligent control method for an unmanned intelligent rotating table, which captures the joint motion characteristics of a biological target through a detection device, and uses programmable metasurface to dynamically control beams to achieve precise tracking. The system uses a spatiotemporal graph convolution network to construct a motion topology graph, analyzes the abnormal behavior patterns of the target, and matches the threat rules library to generate a threat assessment. Based on the assessment result, a control instruction is automatically generated to drive the rotating table to execute lock tracking. The method realizes the full-process automation from target detection, behavior identification to intelligent tracking, significantly improving the intelligent level and response efficiency of perimeter protection. The intelligent remote control is used for an unmanned intelligent rotating table. Through intelligent beam regulation, dynamic topology behavior identification and threat assessment, the unmanned rotating table realizes high-precision automatic tracking and threat response to biological targets.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent remote control, in particular to an intelligent remote control method and system for an unmanned intelligent turntable. BACKGROUND

[0002] In the field of perimeter protection, accurate identification and tracking of concealed targets are key technical requirements. Traditional passive monitoring methods are difficult to effectively detect biological targets (such as hidden personnel, wild animals, etc.) with camouflage or low observable characteristics, especially in complex environments (such as jungles, night or bad weather), existing systems often miss detection or misjudgment. Therefore, an unmanned solution that can actively detect, intelligently identify biological features, and achieve autonomous tracking is urgently needed to improve the real-time performance and concealed target detection rate of security systems.

[0003] Currently, some advanced systems use a fusion scheme of "millimeter wave radar + deep learning", which acquires target micro-motion features such as breathing and joint movement by actively transmitting signals with a millimeter wave radar, and analyzes the reflected signals with a convolutional neural network to realize biological target detection in non-line-of-sight conditions. To some extent, this scheme solves the limitations of optical sensors in concealed scenes and can distinguish biological and non-biological targets through motion features.

[0004] However, this scheme still has significant shortcomings: first, the fixed beam coverage of the millimeter wave radar is limited, making it difficult to dynamically optimize the detection direction, resulting in a lag in tracking fast-moving or intermittently appearing targets; second, the convolutional neural network model only extracts local spatiotemporal features and cannot model the topological correlation of biological joint movements, such as limb coordination during walking, resulting in insufficient behavior recognition accuracy; third, there is a lack of closed-loop linkage with turntable control, relying on manual intervention to adjust the monitoring angle, making it difficult to achieve true autonomous response. SUMMARY

[0005] The present application provides an intelligent remote control method and system for an unmanned intelligent turntable to solve the problem of poor intelligent remote control effect in the prior art.

[0006] In a first aspect, the present application provides an intelligent remote control method for an unmanned intelligent turntable, comprising:

[0007] When the detection device integrated in the unmanned intelligent turntable scans the perimeter area, it captures the reflection signal of the biological target and parses the feature data generated by the subtle movement of the biological joints from the reflection signal;

[0008] Based on the orientation information and motion trajectory of the biological target reflected by the feature data, a beam control instruction is generated;

[0009] Reconstructing an electromagnetic wave phase distribution of the programmable metasurface by responding to the beam steering instruction, and adjusting a direction and a coverage range of the transmission beam of the detection device to enhance a continuous tracking capability of the biological target;

[0010] Inputting feature data acquired by the enhanced continuous tracking capability into a pre-constructed spatio-temporal graph convolution network, constructing a spatial topology node according to a biological joint connection relationship through the spatio-temporal graph convolution network, and mapping time series motion features in the feature data to the spatial topology node to form a motion topology graph;

[0011] Identifying a specific abnormal behavior pattern of the biological target in the perimeter region through the motion topology graph, matching the specific abnormal behavior pattern with a perimeter protection threat rule base to generate a threat level control vector;

[0012] Calculating a trajectory prediction parameter of the biological target based on the threat level control vector, generating a spatial angle control signal of an unmanned intelligent turntable to drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

[0013] Optionally, calculating a trajectory prediction parameter of the biological target based on the threat level control vector, generating a spatial angle control signal of an unmanned intelligent turntable to drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal, includes:

[0014] Determining a trajectory prediction time window length and a spatial trajectory expansion radius according to a threat type code and a threat intensity value in the threat level control vector;

[0015] Taking a spatial coordinate of a biological target at a previous moment as a reference point, generating a main prediction path by extending the trajectory prediction time window length along a motion direction, and generating a fan-shaped prediction area with the spatial trajectory expansion radius as a scattering radius;

[0016] Discretizing the main prediction path into a sequence of spatial coordinate points, calculating an azimuth angle and a pitch angle of each coordinate point relative to the unmanned intelligent turntable and a steering angular velocity between adjacent coordinate points, and generating a spatial angle control signal containing an azimuth angle sequence, a pitch angle sequence, and a steering angular velocity sequence;

[0017] Inputting the spatial angle control signal into a servo controller of the unmanned intelligent turntable, continuously adjusting a pointing azimuth angle and a pitch angle of the turntable according to the steering angular velocity, and triggering an adaptive compensation mechanism when an actual pointing direction of the turntable deviates from a target coordinate point by more than a fault tolerance angle, to calculate a compensation angular acceleration using the actual deviation angle as an input and superimpose the compensation angular acceleration on the steering angular velocity to update the steering parameter.

[0018] Optionally, the spatial angle control signal is input into a servo controller of the unmanned intelligent turntable, and the pointing azimuth and elevation angle of the turntable are continuously adjusted according to the steering angular velocity; when the actual pointing direction of the turntable deviates from the target coordinate point by more than a fault tolerance angle, an adaptive compensation mechanism is triggered, and a compensation angular acceleration is calculated using the actual deviation angle as an input and is superimposed on the steering angular velocity to update the steering parameters, including:

[0019] The spatial angle control signal is input into a servo controller, and the servo controller drives the rotating mechanism of the turntable to continuously adjust the pointing azimuth and elevation angle of the turntable according to the current value in the steering angular velocity sequence, so that the turntable is directed towards the current target position indicated by the azimuth angle sequence and the elevation angle sequence;

[0020] During the adjustment process, the actual azimuth angle and the actual elevation angle of the turntable are obtained, and a first deviation angle between the actual azimuth angle and the current value in the azimuth angle sequence and a second deviation angle between the actual elevation angle and the current value in the elevation angle sequence are calculated.

[0021] The sizes of the first deviation angle and the second deviation angle are compared with a preset fault tolerance angle, and when either the first deviation angle or the second deviation angle exceeds the fault tolerance angle, an adaptive compensation mechanism is activated.

[0022] In the adaptive compensation mechanism, the first deviation angle and the second deviation angle are used as inputs to calculate a first compensation angular acceleration and a second compensation angular acceleration, respectively.

[0023] The first compensation angular acceleration is superimposed on the azimuth angle component of the current value in the steering angular velocity sequence to form an updated azimuth angle steering angular velocity, and the second compensation angular acceleration is superimposed on the elevation angle component of the current value in the steering angular velocity sequence to form an updated elevation angle steering angular velocity, and the updated azimuth angle steering angular velocity and the updated elevation angle steering angular velocity are combined as a whole to update the steering parameters.

[0024] The original corresponding values in the steering angular velocity sequence are replaced by the whole updated steering parameters, and the continuous adjustment process of the pointing direction of the turntable is continued.

[0025] Optionally, the feature data obtained by the enhanced continuous tracking capability is input into a pre-constructed spatio-temporal graph convolution network, and a spatial topology node is constructed according to the biological joint connection relationship through the spatio-temporal graph convolution network, and the time series motion features in the feature data are mapped to the spatial topology node to form a motion topology graph, including:

[0026] transmit the enhanced feature data to a data receiving interface of a spatio-temporal graph convolution network, and create a spatial node array according to the natural connection structure of the biological joints, each spatial node corresponding to the physical position of a specific biological joint and establishing a connection edge between adjacent nodes in anatomy;

[0027] assign the motion parameter time series of each biological joint in the feature data to the time series feature container of the corresponding spatial node;

[0028] combine the spatial node array, the connection edge, and the loaded time series feature container to generate a motion topology graph that integrates the spatial topology structure and the time motion features.

[0029] Optionally, the specific abnormal behavior pattern of the biological target in the perimeter area is identified through the motion topology graph, and the specific abnormal behavior pattern is matched with the perimeter protection threat rule library to generate a threat level control vector, including:

[0030] traverse the spatial nodes of the motion topology graph, extract the fluctuation extreme value and fluctuation frequency of the directional angle change sequence, calculate the burst acceleration number and acceleration amplitude of the rate change sequence, and count the abnormal oscillation period of the amplitude change sequence, and mark the node as an abnormal node if any of the following conditions is met: the directional angle fluctuation frequency exceeds a first set threshold, or the burst acceleration number exceeds a second set threshold in a unit time, or the abnormal oscillation period is shorter than a third set threshold;

[0031] construct an abnormal behavior pattern graph according to the spatial distribution position of the abnormal node and its connection edge relationship, compare the abnormal behavior pattern graph with the rule entries in the perimeter protection threat rule library one by one, and when the similarity between the abnormal behavior pattern graph and any rule entry exceeds a matching threshold, generate a threat level control vector containing a threat type code and a threat intensity value.

[0032] Optionally, when the detection device integrated with the unmanned intelligent turret scans the perimeter area, the reflection signal of the biological target is captured, and the feature data generated by the subtle motion of the biological joints is parsed from the reflection signal, including:

[0033] emit a continuous electromagnetic scanning signal to the perimeter area through the detection device, receive the mixed signal reflected by the biological target, and separate the modulation signal component generated by the periodic micro-displacement of the biological joints from the mixed signal;

[0034] perform time domain segmentation and frequency domain energy focusing on the modulation signal component to extract the oscillation frequency, amplitude fluctuation, and phase offset of the signal waveform in each segmented period;

[0035] The oscillation frequency is mapped to a joint swing rate, the amplitude fluctuation is mapped to a joint displacement amplitude, and the phase shift is mapped to a joint movement direction angle, to generate characteristic data composed of the rate, the amplitude, and the direction angle.

[0036] Optionally, the programmable metasurface is reconfigured in response to the beam steering instruction to adjust the direction and coverage of the emission beam of the detection device to enhance the continuous tracking capability of the biological target, including:

[0037] The target azimuth angle, the target elevation angle, and the spatial cone angle range are parsed from the beam steering instruction.

[0038] According to the target azimuth angle and the target elevation angle, a phase compensation value required to be generated by each control unit of the programmable metasurface is calculated, the phase compensation value causes the electromagnetic waves emitted by each control unit to form coherent superposition in the direction defined by the target azimuth angle and the target elevation angle, and a gradient distribution of the phase compensation value is adjusted according to the spatial cone angle range, so that the width of the coherent superposition beam is expanded to the spatial cone angle range.

[0039] The phase compensation value is written into the control unit array of the programmable metasurface, so that the emission beam forms a dynamic tracking area centered on the target azimuth angle and the target elevation angle and covering the spatial cone angle range, and the dynamic tracking area continuously updates the phase distribution as the biological target moves.

[0040] Optionally, the beam steering instruction is generated based on the azimuth information and the movement trajectory of the biological target reflected by the characteristic data, including:

[0041] According to the direction angle change amount of the continuous time stamp in the characteristic data, a movement vector of the biological target in the three-dimensional space is calculated, and a movement path in a future time window is extrapolated based on the movement vector.

[0042] A position point on the movement path is converted into a spherical coordinate system coordinate with the unmanned intelligent turntable as the origin, to generate a trajectory coordinate sequence including a distance value, an azimuth angle, and an elevation angle.

[0043] According to the maximum offset distance of adjacent coordinate points in the trajectory coordinate sequence, a spatial cone angle range to be covered by the emission beam is determined.

[0044] The azimuth angle, the elevation angle, and the spatial cone angle range of the target tracking point are encapsulated as instruction parameters of the beam steering instruction.

[0045] In a second aspect, the present application provides an intelligent remote control system for an unmanned intelligent turntable, including:

[0046] The first generation module captures the reflection signal of the biological target when the detection device integrated with the unmanned intelligent turntable scans the perimeter area, and analyzes the characteristic data generated by the fine movement of the biological joint from the reflection signal.

[0047] The second generation module generates a beam control instruction based on the position information and movement trajectory of the biological target reflected by the characteristic data.

[0048] The enhancement module reconstructs the electromagnetic wave phase distribution of the programmable metasurface by responding to the beam control instruction, adjusts the direction and coverage range of the transmission beam of the detection device, and enhances the continuous tracking ability of the biological target.

[0049] The formation module inputs the characteristic data obtained by the enhanced continuous tracking ability into a pre-constructed spatio-temporal graph convolution network, constructs a spatial topology node according to the biological joint connection relationship through the spatio-temporal graph convolution network, and maps the time series movement features in the characteristic data to the spatial topology node to form a movement topology graph.

[0050] The third generation module identifies a specific abnormal behavior pattern of the biological target in the perimeter area through the movement topology graph, matches the specific abnormal behavior pattern with a perimeter protection threat rule library, and generates a threat level control vector.

[0051] The driving module calculates the trajectory prediction parameter of the biological target based on the threat level control vector, generates a spatial angle control signal of the unmanned intelligent turntable, and drives the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

[0052] In a third aspect, the embodiments of the present application provide a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the intelligent remote control method for the unmanned intelligent turntable as described in the first aspect above.

[0053] The application embodiment, when the detection device integrated with the unmanned intelligent rotary table scans the perimeter area, captures the reflection signal of the biological target, and analyzes the characteristic data generated by the biological joint fine movement from the reflection signal; based on the position information and motion trajectory of the biological target reflected by the characteristic data, a beam control instruction is generated; the programmable metasurface responds to the beam control instruction, reconstructs the electromagnetic wave phase distribution of the programmable metasurface, and adjusts the direction and coverage range of the emission beam of the detection device, so as to enhance the continuous tracking ability of the biological target; the characteristic data obtained by the enhanced continuous tracking ability is input into the pre-constructed spatio-temporal graph convolution network, so as to construct a spatial topology node according to the biological joint connection relationship through the spatio-temporal graph convolution network, and map the time sequence motion features in the characteristic data to the spatial topology node to form a motion topology graph; through the motion topology graph, a specific abnormal behavior mode of the biological target in the perimeter area is identified, and the specific abnormal behavior mode is matched with a perimeter protection threat rule library to generate a threat level control vector; based on the threat level control vector, a trajectory prediction parameter of the biological target is calculated, a spatial angle control signal of the unmanned intelligent rotary table is generated, so as to drive the unmanned intelligent rotary table to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

[0054] The application has the following beneficial effects:

[0055] The biological joint micro-movement characteristics in the reflection signal are analyzed by the detection device, breaking through the limitations of traditional optical monitoring, and realizing non-contact detection of hidden targets. The programmable metasurface is used to reconstruct the beam phase distribution in real time, realize intelligent focusing and adaptive adjustment of the scanning range of the emission beam, and improve the robustness of target tracking in complex environments. The joint motion features are mapped to a dynamic topology graph through the spatio-temporal graph convolution network, a space-time correlation model of biological motion is established, and the biological rationality of behavior recognition is significantly improved. Based on the matching of the motion topology graph and the threat rule library, the abnormal behavior is quantitatively graded, and an interpretable threat level control vector is generated. According to the threat level, the target trajectory is dynamically predicted and a high-precision spatial angle control signal is generated, realizing the full-process automation of the rotary table from "discovery-identification-locking".

[0056] Further, the application improves the trajectory prediction and tracking control mechanism, dynamically determines the prediction time window and the expansion radius according to the threat level, generates a precise control signal containing the azimuth / elevation angle sequence and the steering angular velocity, and introduces a deviation triggered adaptive compensation mechanism: when the actual pointing deviation is out of limit, the compensation angular acceleration is calculated in real time and the steering parameters are updated, forming a closed-loop control of "prediction-execution-feedback-compensation". The servo controller realizes smooth tracking by continuously adjusting the steering angular velocity, and the compensation mechanism ensures the tracking stability in dynamic environment.

[0057] The application solves the problems of trajectory prediction lag and mechanical response error accumulation of fast moving targets, greatly improves tracking accuracy through adaptive compensation, avoids invalid jitter through fault tolerance angle mechanism, realizes centimeter-level tracking stability under complex motion mode, and reduces the energy consumption of the servo system.

[0058] These aspects or other aspects of the application will be more apparent in the following description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0060] Figure 1 A flow chart of an intelligent remote control method for an unmanned intelligent turntable is shown;

[0061] Figure 2 A structural schematic diagram of an intelligent remote control system for an unmanned intelligent turntable is shown;

[0062] Figure 3 A structural schematic diagram of a computing device is shown. DETAILED DESCRIPTION

[0063] In order to make those skilled in the art better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0064] In some of the descriptions in the specification and claims of the present application and the above-mentioned drawings, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed or in parallel without the order in which they appear in this text. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and the operations can be executed in sequence or in parallel. It should be noted that the "first", "second", etc. in this text are used to distinguish different messages, devices, modules, etc., and do not represent the order of sequence, nor do "first" and "second" represent different types.

[0065] In the current perimeter protection system, the concealed target detection scheme based on millimeter wave radar and CNN has three key defects: first, the fixed beam mode leads to detection blind area, and it is difficult to dynamically track fast-moving or intermittent biological targets; second, the CNN model lacks the modeling ability of biological joint motion topology correlation, resulting in insufficient behavior recognition accuracy; third, the system response has open-loop delay, and manual intervention is needed to adjust the monitoring angle, which cannot realize real autonomous threat response. These problems are essentially caused by the fragmented processing of the three links of "beam control-behavior understanding-execution control" in the prior art, and the lack of unified modeling ability of biological motion spatiotemporal features.

[0066] In view of the above limitations, the present application proposes an intelligent turntable control method based on programmable metasurface and spatiotemporal graph convolution network, which realizes technical breakthrough through the construction of a closed-loop system of "dynamic perception-intelligent analysis-autonomous response". First, the programmable metasurface is used to reconstruct the beam phase distribution in real time, realizing intelligent focusing of the transmitting beam and adaptive adjustment of the scanning range, solving the problem of tracking blind area of fixed beam; second, the spatiotemporal graph convolution network is used to map the joint motion features to a dynamic topology graph, establishing a spatiotemporal correlation model of biological motion, and improving the behavior recognition accuracy; finally, based on the threat level control vector, a high-precision spatial angle signal is generated, and an adaptive compensation mechanism is introduced, realizing the full-process automation of the turntable from target discovery to locking tracking. This scheme improves the concealed target detection rate, reduces the response delay, and completely solves the three core defects of discontinuous detection, inaccurate recognition and untimely response in the prior art.

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0068] Figure 1 A flowchart of an intelligent remote control method for an unmanned intelligent turntable is provided for the embodiments of the present application, as shown in Figure 1 The method comprises:

[0069] 101、When the detection device integrated in the unmanned intelligent turntable scans the perimeter area, the reflection signal of the biological target is captured, and the feature data generated by the subtle movement of the biological joints is analyzed from the reflection signal;

[0070] Optionally, in step 101, when the detection device integrated in the unmanned intelligent turntable scans the perimeter area, the reflection signal of the biological target is captured, and the characteristic data generated by the subtle movement of the biological joint is analyzed from the reflection signal, including:

[0071] 1011. The detection device emits a continuous electromagnetic scanning signal to the perimeter area, receives the mixed signal reflected by the biological target, and separates the modulation signal component generated by the periodic micro-displacement of the biological joint from the mixed signal;

[0072] 1012. Time domain segmentation and frequency domain energy focusing are performed on the modulation signal component to extract the oscillation frequency, amplitude fluctuation and phase offset of the signal waveform in each segmented period;

[0073] 1013. The oscillation frequency is mapped to the joint swing rate, the amplitude fluctuation is mapped to the joint displacement amplitude, and the phase offset is mapped to the joint movement direction angle to generate characteristic data composed of rate, amplitude and direction angle.

[0074] In the above scheme, the perimeter area refers to the boundary range monitored by the unmanned intelligent turntable, for example, 50 meters outside the garden fence. The biological target refers to a moving object with life characteristics, such as a person or an animal. The reflection signal refers to the signal reflected by the electromagnetic wave emitted by the detection device after encountering the biological object. The characteristic data refers to the joint movement parameters extracted from the reflection signal, such as rate, amplitude and direction angle, which are used to identify the biological action pattern. The frequency range of the scanning signal covers the reflection sensitive frequency band of the biological object.

[0075] In the embodiments of the present application, first, the detection device emits a continuous electromagnetic scanning signal to the perimeter area, for example, a millimeter wave with a frequency of When the signal encounters a biological target, the periodic micro-displacement of the joint will cause the reflection signal to produce a Doppler shift, forming a modulation signal component. The mixed signal s(t) obtained at the receiving end can be represented as:

[0076]

[0077] wherein, is the static background reflection amplitude, is the joint reflection signal amplitude, is the Doppler shift frequency, determined by the joint movement speed , , is the speed of light; is the phase offset, is the environmental noise. According to the mixed signal, the static background reflection signal is used as the reference noise, and the filter coefficient is dynamically trained by the LMS algorithm to iteratively subtract the environmental static interference from the mixed signal. The calculation formula is: wherein is a filter coefficient vector, is a step factor, is a reference noise vector. The mixed signal with ambient interference removed is then passed through a 1-5Hz band-pass filter, such as the frequency band characteristic of human joint motion, to filter out residual high-frequency noise, such as mechanical vibration, and low-frequency interference, such as slowly moving vegetation, and finally output a pure biological joint modulation signal. The calculation formula is:

[0078]

[0079] wherein, , is a filter coefficient, , is a filter order.

[0080] Secondly, the separated modulation signal is time-domain segmented by step 1012, for example, every 0.5 seconds is an analysis window, and a Hanning window function is applied in each window to suppress spectral leakage, is the current sampling point sequence number, the value range , is the total number of sampling points in a single period, determined by the sampling rate and the window length , the calculation formula is , for example, for a 50-sample-point signal, the sampling rate , to suppress truncation effects, and generate a windowed signal , is the original signal amplitude value of each sampling point. Perform point FFT to get complex spectrum , wherein is the frequency index, , is the Fourier base function, corresponding to the physical frequency , is the sampling rate, which determines the maximum analysis frequency . Then calculate the amplitude spectrum , locate the maximum value index , then extract the oscillation frequency as , wherein is the cosine term energy, is the sine term energy. Within the half-power bandwidth of , i.e. , the frequency range where the peak value drops to, calculate the amplitude fluctuation , wherein is the -3dB bandwidth boundary index, determined by interpolation method; finally calculate the phase angle at the main frequency , calculate the phase difference between adjacent time periods according to the phase angle , characterizing the change in direction of movement.

[0081] Finally, the oscillation frequency of the signal main frequency is directly corresponded to the flexion and extension frequency of the joint per unit time through step 1013, for example, detecting a 2Hz oscillation frequency represents that the joint completes 2 complete swings per second, which is mapped as the swing rate; Similarly, the amplitude fluctuation quantity is converted into physical displacement through the pre-set calibration curve, for example, a 0.3V voltage difference corresponds to a 5cm flexion amplitude of an adult knee joint through laboratory calibration, and the calibration is based on the inverse square relationship between the electromagnetic wave reflection intensity and the target displacement distance; Similarly, the phase shift quantity is converted into the spatial direction angle according to the kinematic geometric model, for example, a 0.6 rad phase difference is mapped as a right turn of 60°, and the conversion principle is that the phase difference reflects the tangential velocity component of the joint relative to the radar line of sight, and a direction solving model is established combined with the Doppler effect. Finally, the standardized feature data triplets are generated .

[0082] For example, in the perimeter protection scene of region A:

[0083] Specifically, the unmanned turntable scans at a speed of 10 revolutions per minute. When the detection device detects that the B biological target approaches the fence, the millimeter wave signal reflected by the leg joint of the biological target is captured. The 1.5Hz modulation component is separated out by filtering, wherein the corresponding step frequency is 0.5 seconds, and the analysis obtains: the amplitude fluctuation of the first time period is 0.4V (which is mapped to a displacement of 6cm), the phase shift of the second time period is 40° (which is mapped to a right turn of 60°), and finally the feature data [1.5, 6, 60] is generated and marked as "walking and right turning".

[0084] This step realizes the fine analysis of the joint movement of the biological target, converts the original electromagnetic signal into quantifiable behavior feature data, provides high-precision input for subsequent target behavior recognition, and significantly improves the capture ability of potential threat actions in perimeter monitoring.

[0085] 102、based on the position information and the movement trajectory of the biological target reflected by the feature data, generate a beam control instruction;

[0086] Optionally, the step 102 of generating a beam control instruction based on the position information and the movement trajectory of the biological target reflected by the feature data can specifically include:

[0087] 1021, calculate the movement vector of the biological target in the three-dimensional space according to the direction angle change quantity of the continuous time stamp in the feature data, and extrapolate the movement path in the future time window based on the movement vector;

[0088] 1022、Convert the position points on the motion path into spherical coordinate system coordinates with the unmanned intelligent turntable as the origin, to generate a trajectory coordinate sequence containing distance values, azimuth angles, and elevation angles;

[0089] 1023、Determine the spatial cone angle range that the emitted beam needs to cover according to the maximum offset distance of adjacent coordinate points in the trajectory coordinate sequence;

[0090] 1024、Encapsulate the azimuth angle, elevation angle, and spatial cone angle range of the target tracking point as instruction parameters of the beam control instruction.

[0091] In the above scheme, the azimuth information refers to the direction angle (0°-360°) of the biological target relative to the unmanned intelligent turntable.

[0092] The motion trajectory refers to the spatial path of the target moving over time (for example, turning from 30° north to 60° northeast).

[0093] The beam control instruction refers to the instruction parameters for controlling the direction and coverage range of the electromagnetic wave beam emitted by the turntable, including the azimuth angle, the elevation angle, and the cone angle. The feature data refers to the real-time azimuth and motion trajectory of the target.

[0094] In the embodiments of the present application, first, the instantaneous speed is calculated according to the adjacent time stamps in the feature data in step 1021, for example, the direction angle at t=1s is , the direction angle at t=2s is , and the calculation formula is , where is the change amount of the direction angle between consecutive time stamps, for example, , is the time interval; a three-dimensional motion model is established in combination with the historical trajectory points, where the horizontal displacement is , and the vertical displacement is , where is the target distance measured by the radar, is the current azimuth angle, and the movement vector of is calculated according to the horizontal displacement and the vertical displacement, that is, the difference between the horizontal displacement and the vertical displacement of the next second and the previous second is taken as the movement vector. The future path is predicted according to the movement vector, and the formula is , where is the current position coordinate of the target, is the prediction length.

[0095] Secondly, the position points on the predicted future motion path, for example, are converted into spherical coordinates: , where is the elevation angle, is the rectangular coordinate system coordinate, and the calculation results are generated into a trajectory coordinate according to the time sequence: .

[0096] Next, the straight-line distance between adjacent spherical coordinate points is calculated by step 1023 according to the formula , is the distance value between the two adjacent points, is the azimuth angle difference value, and the maximum value of all is taken , for example, 2.5 m, combined with the target speed to calculate the beam coverage radius , wherein the safety factor is 1.2, and finally the spatial cone angle is calculated , wherein is the average distance of the trajectory sequence.

[0097] Finally, the latest position point is extracted from the trajectory sequence as the tracking point by step 1024, for example, t=5s coordinates , and the key parameters of the target tracking point, azimuth, elevation, and cone angle, are structured and packaged as control instructions.

[0098] For example, in a perimeter security scene, an unmanned intelligent turntable detects continuous feature data of a biological target (such as a person):

[0099] Specifically, first, based on the continuous feature data, the movement vector of the biological target is calculated: t=1s direction angle 30°, t=2s direction angle 45°, the change in direction angle second and the distance increment (15m→18m), the horizontal velocity vector [east 2.5m / s, north 1.8m / s] is derived, and the future position is extrapolated (t=3s: east-north 58° direction 22m; t=4s: east-north 62° direction 25m). Next, the predicted rectangular coordinate position is converted to spherical coordinates: t=3s point (18.6m, 11.2m)→ distance , azimuth , elevation (target height difference 1.2m), t=4s point (21.8m, 13.1m)→ spherical coordinates , generating a trajectory sequence [(22, 58, 4.2), (25, 62, 4.0)]. Then, the beam coverage range is calculated according to the trajectory sequence: the maximum displacement distance between adjacent points is calculated: t=3→4s displacement , adding a safety factor of 1.2 to get a beam coverage radius of 4.32m, and calculating the cone angle according to the average distance of 23.5m: . Finally, the latest tracking point is selected to generate a structured instruction.

[0100] This step enables the unmanned turret to have the ability of "prediction-tracking-adaptation", and realizes the continuous locking of the fast-moving target in perimeter protection, effectively solves the problem of tracking loss caused by the sudden turning or acceleration of the target in the traditional system, and greatly improves the reliability of security response.

[0101] 103. Reconstructing the electromagnetic wave phase distribution of the programmable metasurface and adjusting the direction and coverage range of the transmission beam of the detection device by responding to the beam regulation instruction through the programmable metasurface, to enhance the continuous tracking capability of the biological target;

[0102] Optionally, the step 103 of reconstructing the electromagnetic wave phase distribution of the programmable metasurface and adjusting the direction and coverage range of the transmission beam of the detection device by responding to the beam regulation instruction through the programmable metasurface, to enhance the continuous tracking capability of the biological target, comprises:

[0103] 1031. Parsing the target azimuth angle, target elevation angle and spatial cone angle range from the beam regulation instruction;

[0104] 1032. Calculating the phase compensation value required to be generated by each control unit of the programmable metasurface according to the target azimuth angle and the target elevation angle, so that the electromagnetic waves emitted by each control unit form coherent superposition in the direction defined by the target azimuth angle and the target elevation angle, and adjusting the gradient distribution of the phase compensation value according to the spatial cone angle range, so that the width of the coherent superposition beam is expanded to the spatial cone angle range;

[0105] 1033. Writing the phase compensation value into the control unit array of the programmable metasurface, so that the transmission beam forms a dynamic tracking area centered on the target azimuth angle and the target elevation angle and covering the spatial cone angle range, and the dynamic tracking area continuously updates the phase distribution as the biological target moves.

[0106] In the above scheme, the programmable metasurface refers to a smart panel composed of controllable units, and the direction of the beam is changed by adjusting the phase of the units. The phase compensation value refers to the phase offset that each unit needs to adjust, so that the beam focuses in the target direction. The dynamic tracking area refers to an adaptive beam coverage area with the target position as the center and the cone angle as the width.

[0107] In the embodiments of the present application, first, the beam regulation instruction is disassembled according to the structure in step 1031, that is, bytes 0-3: 32-bit floating-point number stores the target azimuth angle, bytes 4-7: 32-bit floating-point number stores the target elevation angle, bytes 8-11: 32-bit floating-point number stores the spatial cone angle range, and bytes 12-15: marks the time stamp of the instruction generation time. According to the field separation, the target azimuth angle, the target elevation angle and the spatial cone angle range are extracted from the instruction pool.

[0108] Secondly, the programmable metasurface forms a focused beam pointing to the target direction (azimuth angle θ, elevation angle φ) and expands to the specified cone angle according to step 1032 , calculates the basic focusing phase, according to the principle of electromagnetic wave propagation, each unit The phase value to be compensated at position ( , ) is , where is the wavelength of the electromagnetic wave, and the compensation makes all units emit waves that are coherently superimposed in the target direction; then expand the beam width, according to the spatial cone angle range , superimpose the secondary phase gradient , where the curvature coefficient is determined by the pre-stored mapping table, for example = 21° corresponds to = -700 rad / m², and the value of controls the degree of beam divergence; then combine the two to get the total phase and normalize to expand the width of the coherently superimposed beam to the interval; finally, convert to a voltage signal and write to the metasurface unit.

[0109] Finally, through step 1033, the target azimuth angle, elevation angle and cone angle parameters in the beam control instruction are converted into a phase distribution matrix; then the phase compensation value is written to the physical address of the metasurface control unit array, that is, each unit receives an independent voltage drive signal, for example, 3.7V corresponds to a 30.52 radian phase shift, and the wavefront is reconstructed by changing the electromagnetic properties of the unit substrate; within a millisecond response period, the array cooperates to generate a focused beam centered on the target azimuth angle and elevation angle, for example, azimuth 62° and elevation 4.0°, while expanding the beam width to the specified cone angle range, for example, 21° cone angle covering the target motion prediction area; the system continuously monitors the target motion trajectory, and when the characteristic data is updated, for example, the target suddenly turns right to change the azimuth angle to 65°, the new phase distribution is calculated according to the process of step 1032 and the control unit is rewritten to realize real-time migration of the beam center and the cone coverage area; in the interference scene, for example, a metal reflection appears in the side lobe direction, an anti-interference mode is automatically triggered: the cone angle is contracted to 18° to improve the main lobe gain, while the side lobe suppression ability is enhanced, ensuring stable tracking with a signal-to-noise ratio > 15dB under the conditions of target acceleration, turning or occlusion.

[0110] For example, in the perimeter protection area A, the unmanned turntable detects the target personnel C (height 1.75m) moving at a speed of 1.8m / s northeast, and the initial position is 25m away from the turntable (azimuth angle 62°, elevation angle 4.0°). The system has generated a beam control instruction: target azimuth angle , target elevation angle spatial cone angle .

[0111] Specifically, through the formula Compensation for path difference ensures the beam is precisely pointed to the target center; superimposed secondary phase distribution This expands the beamwidth from the original 5° to 21°. Final phase value The signal is converted to a voltage signal, and the converted 30.52 radians → 3.7V is written to the control unit. The reconstructed beam completes dynamic tracking within 0.6 seconds, i.e., the initial state (t=0 seconds): beam center locked at 62° / 4.0°, -3dB power point width 21.2°, sidelobe suppression <-20dB; target turning (t=0.5 seconds): personnel C suddenly turns right and accelerates to 2 meters per second. The system extrapolates that its position will reach an azimuth angle of 65°, immediately updates the command parameters (center 65°, cone angle 22°), and recalculates and writes the new phase distribution within 18 milliseconds; anti-interference response (t=1.2 seconds): when the metal fence generates strong reflection interference in the beam sidelobe direction, the system automatically increases the phase gradient curvature coefficient, shrinks the beam width to 18°, strengthens the sidelobe suppression to <-25dB, and maintains the target signal signal-to-noise ratio >15dB.

[0112] This step ensures that targets are continuously tracked even when they are moving rapidly or obstructed, significantly improving monitoring stability in complex environments by dynamically reconstructing high-precision directional beams.

[0113] 104. Input the feature data obtained under the enhanced continuous tracking capability into a pre-constructed spatiotemporal graph convolutional network, so as to construct spatial topology nodes based on the biological joint connection relationship through the spatiotemporal graph convolutional network, and map the time series motion features in the feature data to the spatial topology nodes to form a motion topology graph;

[0114] Optionally, in step 104, the feature data acquired under the enhanced continuous tracking capability is input into a pre-constructed spatiotemporal graph convolutional network to construct spatial topology nodes based on biological joint connections, and the time-series motion features in the feature data are mapped to the spatial topology nodes to form a motion topology graph, including:

[0115] 1041. The feature data acquired under the enhanced continuous tracking capability is transmitted to the data receiving interface of the spatiotemporal graph convolutional network, and a spatial node array is created according to the natural connection structure of biological joints. Each spatial node corresponds to the physical position of a specific biological joint, and connection edges are established between anatomically adjacent nodes.

[0116] 1042. Assign the time series of motion parameters of each biological joint in the feature data to the temporal feature container of the corresponding spatial node;

[0117] 1043、combine the spatial node array, connection edges and loaded temporal feature containers to generate a motion topology graph that integrates spatial topology and temporal motion features.

[0118] In the above scheme, the feature data refers to the joint motion parameters extracted from the reflection signal, including the time series of motion parameters of multiple biological joints, such as 1.5 times per second swing of the knee, 6 cm amplitude, and 60° direction. The spatio-temporal graph convolution network refers to an artificial intelligence model that can simultaneously analyze the spatial position relationship of the joint, such as the connection between the knee and the hip joint, and the time variation law, such as the process of the knee amplitude changing from 6 cm to 8 cm. The spatial topology node refers to a point representing a biological joint, such as a knee, an elbow, arranged according to the real physiological position, such as the knee node below the hip node. The connection edge refers to the connection between the nodes, representing the anatomical connection relationship (such as the thigh = the connection between the knee node and the hip node). The temporal feature container refers to the "data bucket" attached to each node, used to store the motion parameters of the joint changing over time, such as the knee node storing the swing amplitude [6 cm, 7 cm, 8 cm…] for 5 seconds. The motion topology graph refers to the final generated dynamic relationship graph, including the joint position, connection relationship and historical motion data, such as showing "how the knee and hip joint work together when the person lifts the leg". The time series of motion parameters includes the rate change sequence, the amplitude change sequence and the direction angle change sequence.

[0119] In the embodiments of the present application, first, the enhanced feature data set obtained in the continuous tracking stage is transmitted to the standardized input interface of the spatio-temporal graph convolution network through the high-speed data bus through step 1041. The data set is represented as a four-dimensional tensor , wherein is the time series length, is the number of detected biological joints, is the feature dimension of each joint, such as 3 dimensions of rate, amplitude, and direction angle; at the same time, a pre-set biological anatomical structure template library is loaded. For human targets, a pre-defined skeleton topology mapping function is used, which assigns a standard anatomical position to each joint type, wherein the node coordinates are based on the biological coordinate system, the origin is the center of the hip joint, and the positioning rule is calculated based on the geometric proportional model, such as the position of the knee joint:

[0120]

[0121] , wherein, is the knee height offset constant stored in the anatomical database. Then, the system instantiates the spatial node array, creates the node set as , assigns the node attributes as the physical position and the spatial index , and then according to Connect edge set is established, and connection verification is checked by using an anatomical rule base, i.e., distance constraint wherein is an anatomical length of a limb segment, and angle constraint wherein is an anatomical reference vector, and a structured spatial topology graph is finally output according to the distance constraint and the angle constraint wherein is a node set, is a connect edge set, is a position matrix.

[0122] Secondly, by step 1042, the anatomical identifier is strictly matched with the joint identity according to the constructed spatial topology node array, for example, the "left knee" node only receives feature data with the label "LeftKnee", and three data integrity checks are performed: detecting time stamp continuity to ensure uninterrupted data per second, verifying parameter integrity to confirm that the rate, amplitude, and direction angle three-dimensional data are complete, and checking physiological reasonableness to mark abnormal data, for example, when the knee swing rate exceeds the range of 0.5-3.5 Hz, an alarm is given. After successful matching, the system initializes a dedicated container for each joint node, i.e., a first-in-first-out cache structure with fixed time length storage capacity (default to retain 30 seconds of data), stores the motion parameters in time sequence into the container while automatically associating the spatial coordinate label, for example, the left knee data always carries the position [-0.15, -0.45, 0], and unifies the standardized numerical value format, i.e., converts the angle to 0-360 degree standard value. Dynamic anomaly protection is implemented during loading: the mean value of adjacent values is automatically interpolated for a short interruption within 0.2 seconds, the data is temporarily stored and delayed writing when electromagnetic interference characteristics are detected, and the original signal is re-verified when motion conflict is found, for example, the joint simultaneously records opposite states of flexion and extension; finally, the data pointer is reset to mark the latest state point when loading is completed, a time-motion intensity curve index is generated, and a container storage state report is output, for example, "right elbow container 28 / 30 seconds full", ensuring that all joint motion history from the walking swing of the knee joint to the climbing and grabbing of the wrist joint is archived to the spatio-temporal container matched with the anatomical position with millisecond-level precision, forming a standardized dynamic behavior database.

[0123] Finally, the physical coordinate information of the spatial node array is called by step 1043 to construct the basic skeleton, such as the three-dimensional position relationship of the 15 human body joint nodes, and the connection edges between adjacent nodes are established according to strict anatomical rules, such as the rigid connection between the knee joint node and the hip joint node, and the abnormal connection is removed through distance threshold checking, such as the edge exceeding 120% of the maximum length of the thigh bone; then the loaded time sequence feature container is accurately mounted to the corresponding spatial node, such as the left wrist joint container binding the left wrist node, and the time synchronization engine aligns the time stamps (microsecond level accuracy) of all containers to fuse the static coordinates of each node with the dynamic motion parameters into a four-dimensional state vector, i.e. spatial three-dimensional coordinates + time dimension motion parameters; then the real-time interaction strength of the connection edge is calculated by the dynamic weight engine, i.e. based on the joint angle difference and the relative speed change rate, when the main motion joint is detected, such as the wrist when climbing, the weight of the related edge is automatically strengthened, such as the wrist-elbow edge weight is increased to 0.9; finally, the standardized motion topology graph data structure is generated: including the node set, i.e. the vector group with coordinates and time sequence, the edge set, i.e. the effective anatomical connection with weight, and the time stack, i.e. the topology snapshot of continuous time period, the topology self-optimization is triggered once every 200 milliseconds, the redundant nodes with frozen displacement rate less than 0.01 are frozen, the secondary connection with hidden weight less than the threshold is hidden, and the non-active joint node level is compressed, and the motion topology graph with anatomical accuracy and behavior representation is output.

[0124] For example, perimeter protection area A, unmanned turntable monitors personnel E approaching the fence:

[0125] Specifically, the system creates a spatial topology containing 15 joint nodes (head, shoulder, elbow, knee, ankle, etc.), connected according to human structure (e.g. left knee node -> left hip node, right knee node -> right hip node). Then assign time sequence data to nodes: left knee node container for continuous 3 seconds amplitude data [0.3cm, 5.1cm, 9.2cm] (slowly lifting leg). Right elbow node container for direction angle [30°, 45°, 60°]. According to the above, generate motion topology graph and analyze: left knee amplitude reaches 9.2cm (exceeds threshold 7cm), while left hip node direction angle changes +20° (hip joint twist). Finally output spatio-temporal graph convolution network identifies as "climbing preparation action", triggers alarm.

[0126] This step integrates discrete joint motion data into a dynamic topology graph, allowing the system to understand the overall coordination of complex actions, such as recognizing "running" requires high-frequency leg swing + synchronized arm coordination. Compared with traditional single-point analysis, this method significantly improves the recognition accuracy of continuous threat actions such as climbing, falling, and crawling, while reducing false positives caused by local data fluctuations, such as joint data anomalies caused by wind blowing clothes.

[0127] 105. identifying a specific abnormal behavior pattern of the biological target in the perimeter area through the motion topological graph, and matching the specific abnormal behavior pattern with a perimeter defense threat rule base to generate a threat level control vector;

[0128] Optionally, the step 105 of identifying a specific abnormal behavior pattern of the biological target in the perimeter area through the motion topological graph, and matching the specific abnormal behavior pattern with a perimeter defense threat rule base to generate a threat level control vector can specifically include:

[0129] 1051. traversing all spatial nodes of the motion topological graph, extracting fluctuation extreme value and fluctuation frequency of the direction angle change sequence from the time sequence feature container of each node, calculating burst acceleration number and acceleration amplitude of the rate change sequence, and counting abnormal oscillation period of the amplitude change sequence, and marking a node as an abnormal node if any of the following conditions is met: the fluctuation frequency of the direction angle exceeds a first set threshold value, or the burst acceleration number exceeds a second set threshold value in a unit time, or the abnormal oscillation period is shorter than a third set threshold value;

[0130] 1052. constructing an abnormal behavior pattern graph according to the spatial distribution position of the abnormal node and the connection edge relationship thereof, and comparing the abnormal behavior pattern graph with rule entries in the perimeter defense threat rule base one by one, and when the similarity between the abnormal behavior pattern graph and any rule entry exceeds a matching threshold value, generating a threat level control vector containing threat type coding and threat intensity value.

[0131] In the above scheme, the dynamic motion topological graph refers to a data structure describing the spatial position, connection relationship and motion history of the biological joint. The abnormal behavior pattern refers to an action that violates the normal behavior characteristics, such as rapid climbing and latent squatting. The threat rule base refers to a pre-defined threat behavior characteristic library, such as the "climbing" rule containing large upper limb movement + lower limb support. The threat level control vector refers to the output result. The abnormal behavior pattern graph contains the motion transmission direction between adjacent abnormal nodes and the geometric distribution form of the abnormal node group. The rule entry contains the pre-set illegal climbing node distribution form, the pre-set rapid advance motion transmission direction and the pre-set latent movement oscillation period characteristics. The threat type coding corresponds to the matched rule entry, and the threat intensity value is calculated based on the number of abnormal nodes and the fluctuation amplitude.

[0132] In the embodiments of the present application, first, all spatial nodes of the motion topological graph are traversed by the step 1051, and three-level motion feature extraction and threshold determination are performed on the time sequence feature container of each node: first, the direction angle change sequence is extracted, the number of absolute values of angle differences between adjacent time points exceeding 60° is calculated as the fluctuation frequency, and the maximum angle jump value is recorded as the fluctuation extreme value, for example, the wrist joint sequence in to ​ The jump is the extreme value, and the fluctuation frequency is detected by the formula whether it exceeds the standard:

[0133]

[0134] Wherein 5 times per second is the first set threshold, that is, more than 5 times per second is judged as abnormal, is the time point , the direction angle value, is the indicator function, if the condition is met, it is 1. Then calculate the burst acceleration characteristics of the speed change sequence, and detect the mutation point of the relative change rate of speed more than 150% point by point:

[0135]

[0136] Wherein, 3 times per second is the second set threshold, that is, more than 3 times per second is triggered to mark, is the joint rate at time point ; Finally, the oscillation periodicity of the amplitude change sequence is counted, and the main frequency period value is obtained by fast Fourier transform:

[0137]

[0138] Wherein, 0.1 seconds is the third set threshold, that is, the main period shorter than 0.1 seconds is considered as abnormal oscillation, is the Fourier transform operator, is the amplitude sequence; When any node meets any condition of direction angle fluctuation frequency exceeding limit, burst acceleration number exceeding threshold or oscillation period being too short at the same time, the system automatically marks it as an abnormal node, and records its abnormal feature type with spatial coordinate index, for example, the left knee joint is marked as "acceleration type abnormal node" because the speed change sequence [1.1, 1.2, 4.7, 4.9] rad / s detects 4 times of speed jump > 150% in 0.3 seconds, exceeding the threshold 3 times per second.

[0139] Finally, the topology structure with weights is established according to the node spatial distribution position and connection edge relationship by step 1052, that is, taking the abnormal node as the vertex, for example, the left wrist and right knee node, generating the edge according to the anatomical connection rule, for example, the wrist- elbow- shoulder chain, and assigning a weight value to each edge, the weight calculation method is the product of the total node abnormal value and the distance attenuation coefficient, wherein the wrist node abnormal value is 7.5 times the quantized value of the direction angle fluctuation frequency, the knee node abnormal value is 3.3 times the number of sudden accelerations, and the distance attenuation coefficient is the inverse of the node distance; then the atlas is compared with the perimeter protection threat rule base item by item: extract the key topological features in the rule item, for example, the "climbing behavior" item requires the upper limb wrist / elbow node and the lower limb knee / ankle node to be abnormal at the same time and the spatial distance to be less than 1 meter, calculate the atlas similarity, wherein is the node set matched by the abnormal behavior pattern atlas and the rule item, is the matching node in the abnormal atlas, is all the necessary node set specified by the rule item, is the preset reference weight of the node in the rule item, is the spatial position coincidence degree, for example, the wrist-knee distance in the actual atlas is 0.8 meters, which is less than the rule threshold of 1 meter, so the coincidence degree is 100%; when the similarity exceeds the matching threshold (default 85%), a threat level control vector is generated according to the matching rule item, the threat type code is taken from the rule ID (such as P2 represents climbing), and the threat intensity value is calculated by normalizing the abnormal node weight wherein is the abnormal feature quantized value of the node , for example, wrist (7.5) + knee (3.3) = 10.8 -> normalized intensity 7.2, and the final output is a vector such as [P2, 7.2].

[0140] For example, a suspicious target is found in the perimeter protection area D:

[0141] Specifically, the angle mutation sequence [35°→152°→38°] (2 times of more than 90° mutation within 0.5 seconds), the fluctuation frequency reaches 6Hz (> threshold 5Hz) -> marked abnormal, the rate sequence [1.1, 1.2, 4.7, 4.9] rad / s (4 times of acceleration more than 1.5 times within 0.3 seconds), the number of sudden accelerations is 4 times per second (> threshold 3 times) -> marked abnormal, the amplitude sequence oscillation period is 0.08 seconds (< threshold 0.1 second) -> marked abnormal; the spatial correlation chain of the upper limb (wrist) and lower limb (knee / ankle) nodes of the above abnormal behavior is formed, and then compared with the rule base according to the rule, compared with the "jumping over the fence" rule: the upper limb mutation and the lower limb acceleration must exist at the same time, the node matching degree: wrist / knee / ankle all meet the key points of the rule, the similarity calculation: 92% (> threshold 85%), and the threat vector , Encode the "Climbing over" behavior type in the rule base. Threat intensity value (based on the number of abnormal nodes × intensity coefficient) .

[0142] This step involves joint-level anomaly detection using a dynamic motion topology map. The system can accurately identify hidden threatening behaviors that are difficult to capture with traditional monitoring and quantify the risk level based on biological kinematics rules. When a spatial correlation pattern of high-frequency mutations in the upper limbs and sudden accelerations in the lower limbs is detected, high-risk behavior items such as "climbing" and "jumping" are automatically matched to generate a threat vector containing behavior type and intensity level. This enables intelligent judgment from microscopic joint anomalies to macroscopic behavioral threats, significantly improving the early warning capability for complex threats such as camouflage and rapid attacks, while avoiding misjudgments caused by animal movement or vegetation swaying.

[0143] 106. Calculate the trajectory prediction parameters of the biological target based on the threat level control vector, generate a spatial angle control signal for the unmanned intelligent turntable, and drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

[0144] Optionally, step 106, which calculates the trajectory prediction parameters of the biological target based on the threat level control vector and generates a spatial angle control signal for the unmanned intelligent turntable to drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal, may specifically include:

[0145] 1061. Determine the trajectory prediction time window length and spatial trajectory expansion radius based on the threat type code and threat intensity value in the threat level control vector;

[0146] 1062. Using the spatial coordinates of the biological target at the previous moment as a reference point, the length of the trajectory prediction time window is extended along the direction of motion to generate the main prediction path, and a fan-shaped prediction area is generated using the spatial trajectory extension radius as the scattering radius.

[0147] 1063. Discretize the main predicted path into a sequence of spatial coordinate points, calculate the azimuth and pitch angles of each coordinate point relative to the unmanned intelligent turntable and the turning angular velocity between adjacent coordinate points, and generate a spatial angle control signal containing the azimuth angle sequence, pitch angle sequence and turning angular velocity sequence.

[0148] 1064. Input the spatial angle control signal into the servo controller of the unmanned intelligent turntable, continuously adjust the azimuth and pitch angles of the turntable according to the steering angular velocity, and trigger the adaptive compensation mechanism when the actual direction of the turntable deviates from the target coordinate point beyond the tolerance angle. Calculate the compensation angular acceleration with the actual deviation angle as input and superimpose the compensation angular acceleration onto the steering angular velocity to update the steering parameters.

[0149] In step 1064, the spatial angle control signal is input into the servo controller of the unmanned intelligent turntable, the pointing azimuth and elevation angle of the turntable are continuously adjusted according to the steering angular velocity, and when the actual pointing of the turntable deviates from the target coordinate point by more than the fault tolerance angle, the adaptive compensation mechanism is triggered, the compensation angular acceleration is calculated according to the actual deviation angle, and the compensation angular acceleration is superimposed on the steering angular velocity to update the steering parameters. The process includes: inputting the spatial angle control signal into the servo controller, and driving the rotating mechanism of the turntable to continuously adjust the pointing azimuth and elevation angle of the turntable according to the current value in the steering angular velocity sequence, so that the turntable points to the current target position indicated by the azimuth angle sequence and the elevation angle sequence; during the adjustment process, the actual azimuth angle and the actual elevation angle of the turntable are obtained, and the first deviation angle between the actual azimuth angle and the current value of the azimuth angle sequence and the second deviation angle between the actual elevation angle and the current value of the elevation angle sequence are calculated; compare the size of the first deviation angle and the second deviation angle with the preset fault tolerance angle, and when any one of the first deviation angle or the second deviation angle exceeds the fault tolerance angle, activate the adaptive compensation mechanism; in the adaptive compensation mechanism, the first compensation angular acceleration and the second compensation angular acceleration are calculated respectively by taking the first deviation angle and the second deviation angle as inputs; the first compensation angular acceleration is superimposed on the azimuth component of the current value of the steering angular velocity sequence to form an updated azimuth steering angular velocity, and the second compensation angular acceleration is superimposed on the elevation component of the current value of the steering angular velocity sequence to form an updated elevation steering angular velocity, and the updated azimuth steering angular velocity and elevation steering angular velocity are combined as a whole to update the steering parameters; the whole updated steering parameters are used to replace the corresponding values in the original steering angular velocity sequence, and the continuous adjustment process of the pointing of the turntable is continued.

[0150] In the above scheme, the threat level control vector refers to a two-tuple containing a threat type code (such as P2 = climbing) and a threat intensity value (such as 7.2 / 10)

[0151] The trajectory prediction parameter refers to the estimated parameter of the future motion of the biological target, i.e. the prediction time length and the path radius. The spatial angle control signal refers to the instruction for controlling the rotation of the turntable, i.e. the azimuth angle sequence, the elevation angle sequence, and the steering angular velocity. The adaptive compensation mechanism refers to the automatic correction system when the turntable deviates from tracking.

[0152] In the embodiments of the present application, first, the threat type code in the threat level control vector is used as the input of the threat type recognition model in step 1061 and The trajectory prediction parameter is accurately calculated, and the core calculation model is:

[0153] The prediction time window length is:

[0154]

[0155] wherein, is a type time coefficient, is a threat intensity value, is a base time window.

[0156] Space trajectory expansion radius:

[0157]

[0158] wherein is a type radius coefficient, is a base expansion radius.

[0159] Secondly, by step 1062, with the latest captured biological target space coordinates, for example, azimuth , elevation , distance meters as the reference point, based on historical trajectory analysis of the motion direction, through the first 5 coordinate points, for example, azimuth sequence , elevation sequence fit the direction vector, determine that the target moves northeast by east at a horizontal turning rate of 0.7 degrees per second and an elevation change rate of 0.06 degrees per second; then extend the trajectory prediction window length along this motion direction, for example, 10.99 seconds determined by the destruction class threat intensity 9.3, calculate the displacement increment to form the main prediction path according to the target current moving speed of 2.3 meters per second, 25.28 meters (10.99 x 2.3) axial extension line, the path endpoint coordinates are azimuth (65.2 + 0.7 x 10.99), distance 53.88 meters (28.6 + 25.28), elevation (3.8 - 0.06 x 10.99); finally, with a space trajectory expansion radius of 4.86 meters as the scattering radius, a fan-shaped prediction area with a 60-degree opening angle is developed on both sides of the main prediction path, for example, the left and right boundaries of the fan-shaped area reach azimuth and , covering an east-west span of 8.3 meters and a north-south span of 5.1 meters, this area completely contains the target's possible turning path deviation, for example, the target suddenly turns to azimuth at a distance of 52 meters is still covered, while avoiding the expansion of invalid areas, the fan-shaped radius is constrained not to exceed the terrain boundary.

[0160] Then, by step 1063, first determine the total number of discrete points based on the prediction window length and the time step, and then generate each point position by linear interpolation from the path starting point coordinates to the endpoint coordinates, wherein the coordinate calculation formula of the first point is:

[0161]

[0162] wherein, is the total number of discrete points, for reference azimuth, then calculate the azimuth of each point relative to the turntable and the pitch angle , for the height of the turntable, then calculate the steering angular velocity by central difference method , for the time step, finally generate the synchronization control signal containing the azimuth angle sequence , the pitch angle sequence , and the steering angular velocity sequence .

[0163] Finally, after inputting the spatial angle control signal into the servo controller of the unmanned intelligent turntable through step 1064, first drive the rotating mechanism to continuously adjust the pointing direction of the turntable according to the current value in the steering angular velocity sequence, i.e. the directional angle component and the pitch angle component ; in the process of movement, real-time collection of the actual azimuth and the actual pitch angle of the turntable, respectively, and the target sequence value to calculate the deviation angle ; when any deviation angle exceeds the preset fault tolerance angle , trigger the adaptive compensation mechanism: first calculate the compensation angular acceleration by the proportional differential control formula:

[0164] ,

[0165] where the proportional gain strengthens the deviation correction, the differential gain suppresses sudden disturbances, and the deviation rate represents the disturbance intensity; then add the compensation acceleration to the original steering angular velocity:

[0166]

[0167] where the control period ensures real-time updating, and the angular acceleration is converted into a physical process of velocity increment; finally, replace the original sequence value with the updated steering parameter , and the servo controller immediately drives the turntable to execute the correction movement, forming a "sensing-computing-compensation-execution" closed-loop control. For example, when the crosswind causes an azimuth deviation of 1.2° (change rate 3° / s), by generating a compensation acceleration , the steering speed is increased from 5.0° / s to 0.16° / s, and the deviation is suppressed to 0.2° within 0.3 seconds, maintaining the locking accuracy of azimuth ±0.3° and pitch angle ±0.1°.

[0168] For example, after a biological target (threat type code S2 - armed assault, threat intensity 9.3) is found in the perimeter security zone F:

[0169] Specifically, the prediction window length coefficient 2.0 is determined according to the threat type S2 (assault type), the intensity 9.3 corresponds to the radius coefficient 2.2, the prediction window length = 3 seconds (base value) x 2.0 = 6 seconds, and the spatial trajectory expansion radius = 1.5 meters x 2.2 = 3.3 meters. Set the reference parameters: longer warning is needed for assault-type threats, so both the window length and the radius are doubled, then the previous time target coordinates (azimuth 48.5°, distance 22 meters) are taken as the reference point to extend the main path for 6 seconds along the motion direction (northeast 58°), and the path endpoint is extended 18 meters to the coordinate point (40.2m, 58°) to generate a fan-shaped prediction area: taking the main path as the center line, the scattering area is expanded to ±3.3 meters to the left and right, i.e. the left boundary: azimuth , distance 40m, and the right boundary: azimuth , distance 40m, when the target makes a sharp turn. That is, from sharp turn , the scattering radius ensures that it is within the prediction area; then the main path is discretized into 12 points (0.5 second interval), the 7th point coordinates (32m, ) are calculated → azimuth angle = , target height 1.7 meters → pitch angle = , the 7th→8th point turning angle velocity is calculated: azimuth angle = , 0.5 second interval → angular velocity 1.4° / s, Δ pitch angle = → angular velocity / s, output signal: azimuth angle sequence , pitch angle sequence , angular velocity sequence , finally at t=3.5 seconds, the strong side wind causes the actual azimuth angle (59.8°) to deviate from the target point, with a deviation of 0.7°> tolerance angle 0.5°→ trigger compensation, i.e. proportional compensation term: Kp , differential compensation term: Kd , compensation angular acceleration = , original turning angular velocity / s → updated to , the turntable corrects 0.65° in 0.25 seconds and restores the lock on the target coordinate point.

[0170] This step uses a threat-driven prediction model, which automatically extends the prediction window and expands the tracking range when the target is climbing, running, or engaging in other high-risk behaviors; the coordination control of the coordinate sequence and the angular velocity signal realizes smooth turning of the turntable; when strong wind or sudden change of target direction causes tracking deviation, the compensation mechanism dynamically corrects the angle within 0.2 seconds, ensuring continuous locking on high-speed moving targets and significantly reducing the risk of target loss in complex environments.

[0171] Here are the complete embodiments of steps 101-106:

[0172] In the perimeter protection system of the industrial park, at 2:15 a.m., the unmanned intelligent turret scans the 50-meter protection zone outside the perimeter wall at a speed of 10 revolutions per minute, and detects a suspicious biological target (about 1.75 meters tall) hiding and moving in the C area vegetation shadow.

[0173] First, the turret emits 77GHz millimeter waves to capture the target's leg joint reflection signal. After filtering out the static interference of the vegetation by the LMS algorithm, a 1.8Hz biological modulation signal is separated. Time-frequency analysis is performed with a 0.5 second analysis window: in the first period, the amplitude fluctuation of 0.45V (mapped to a displacement of 6.8cm by the calibration curve) and the phase difference of 0.7rad (calculated as a right turn of 50° by the geometric model) are detected; in the second period, a sudden 2.5Hz high-frequency oscillation (> normal walking threshold of 1.5Hz) is captured. Output feature data triplets: [1.8, 6.8, 50], [2.5, 0.3, 55].

[0174] Second, based on the feature data of consecutive frames (azimuth ), the movement vector is calculated: horizontal speed 2.8m / s (east-north), pitch rate -0.05° / s. Predict the trajectory in the next 5 seconds: the reference point (azimuth , distance 25m) extends to (azimuth , distance 39m). Calculate the beam cone angle: the maximum offset distance of adjacent points is 3.2m x safety factor 1.2 = 3.84m, the average distance is 32m → cone angle . Programmable metasurface reconstructs the beam: phase compensation value , wavelength =3.9mm, the control unit writes 3.8V voltage to make the beam center point to 55° azimuth / 4° pitch, and the width is expanded to 13.7° cone angle to cover the predicted path.

[0175] Next, the spatio-temporal graph convolution network constructs the motion topology graph: the spatial nodes are 15 joints (focus on the left knee node coordinates [-0.15, -0.45, 0]); the time sequence container is the left knee loaded [0.4cm, 5.2cm, 8.1cm] leg lifting data, and the abnormal feature detected is the sudden acceleration frequency of the left knee node 4 times per second (> threshold of 3 times), and the right wrist direction angle mutation rate 7Hz (> threshold of 5Hz). Generate abnormal atlas: knee-wrist spatial distance 1.1m, edge weight 0.78. Match threat rule library: similarity to "crouching and climbing" rule 91% (> threshold of 85%). Output threat vector [L3, 7.8] (L3 = crouching threat, intensity 7.8).

[0176] Finally, the prediction parameters are calculated based on the threat vector: latent class window length coefficient 1.8→T=3s×1.8=5.4s; intensity 7.8→radius factor 1.52→R=1.5m×1.52=2.28m. The discretization main path is 11 points (time step 0.5s), and the 6th point coordinate (azimuth 61.2°, distance 34.5m):

[0177] azimuth angle ;

[0178] pitch angle ;

[0179] angular velocity .

[0180] Strong side wind causes actual azimuth deviation of 1.5° (> fault tolerance angle 0.5° compensation: , update steering speed , 0.28s correction, deviation to 0.2°, and the turntable continues to lock the target azimuth.

[0181] Figure 2 An embodiment of the present application provides a structure schematic diagram of an intelligent remote control system for an unmanned intelligent turntable, as Figure 2 shown, the system comprises:

[0182] A first generation module 21, when the detection device integrated in the unmanned intelligent turntable scans the perimeter area, captures the reflection signal of the biological target, and analyzes the characteristic data generated by the subtle movement of the biological joint from the reflection signal;

[0183] A second generation module 22, based on the azimuth information and motion trajectory of the biological target reflected by the characteristic data, generates a beam control instruction;

[0184] An enhancement module 23, which responds to the beam control instruction through a programmable metasurface, reconstructs the electromagnetic wave phase distribution of the programmable metasurface, and adjusts the direction and coverage range of the transmission beam of the detection device, to enhance the continuous tracking ability of the biological target;

[0185] A formation module 24 inputs the characteristic data obtained under the enhanced continuous tracking ability into a pre-constructed spatio-temporal graph convolution network, so as to construct a spatial topology node through the spatio-temporal graph convolution network according to the biological joint connection relationship, and map the time series motion features in the characteristic data to the spatial topology node, to form a motion topology graph;

[0186] A third generation module 25, which identifies a specific abnormal behavior mode of the biological target in the perimeter area through the motion topology graph, and matches the specific abnormal behavior mode with a perimeter protection threat rule library to generate a threat level control vector;

[0187] The driving module 26 controls the trajectory prediction parameter of the biological target based on the threat level control vector, generates a spatial angle control signal of the unmanned intelligent turntable, and drives the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

[0188] Figure 2 The intelligent remote control system for the unmanned intelligent turntable can perform Figure 1 The intelligent remote control method for the unmanned intelligent turntable of the embodiment described above achieves the principles and technical effects, which will not be described again. The specific operation of each module and unit of the intelligent remote control system for the unmanned intelligent turntable in the above embodiment has been described in detail in the embodiment related to the method, which will not be described in detail here.

[0189] In one possible design, Figure 2 The intelligent remote control system for the unmanned intelligent turntable of the embodiment described above can be implemented as a computing device, such as Figure 3 As shown, the computing device can include a storage component 31 and a processing component 32.

[0190] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0191] The processing component 32 is used for the above Figure 1 The intelligent remote control method for the unmanned intelligent turntable of the embodiment described above.

[0192] The processing component 32 can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components, for executing the above method.

[0193] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0194] Of course, the computing device can also include other components, such as input / output interfaces, display components, communication components, etc.

[0195] The input / output interface provides an interface between the processing component and peripheral interface modules, which can be output devices, input devices, etc.

[0196] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0197] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the processing component, the storage component, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0198] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0199] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0200] From the foregoing description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software and the necessary general hardware platform, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the methods described in each embodiment or some part of the embodiment.

[0201] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent remote control of an unmanned intelligent turntable, characterized in that, include: When the detection device integrated into the unmanned intelligent turntable scans the perimeter area, it captures the reflected signals of biological targets and extracts feature data generated by the subtle movements of biological joints from the reflected signals. Based on the orientation information and movement trajectory of the biological target reflected by the feature data, a beam control command is generated. By responding to the beam control command through a programmable metasurface, the electromagnetic wave phase distribution of the programmable metasurface is reconstructed, and the direction and coverage of the emitted beam of the detection device are adjusted to enhance the continuous tracking capability of the biological target. The feature data acquired under the enhanced continuous tracking capability is input into a pre-constructed spatiotemporal graph convolutional network. The spatiotemporal graph convolutional network constructs spatial topology nodes based on the connection relationship of biological joints, and maps the time-series motion features in the feature data to the spatial topology nodes to form a motion topology graph. The motion topology map is used to identify specific abnormal behavior patterns of the biological target within the perimeter area, and the specific abnormal behavior patterns are matched with the perimeter protection threat rule base to generate a threat level control vector. Based on the threat level control vector, the trajectory prediction parameters of the biological target are calculated, and a spatial angle control signal for the unmanned intelligent turntable is generated to drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

2. The method according to claim 1, characterized in that, Based on the threat level control vector, trajectory prediction parameters of the biological target are calculated, and a spatial angle control signal for the unmanned intelligent turntable is generated to drive the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal, including: The trajectory prediction time window length and spatial trajectory expansion radius are determined based on the threat type code and threat intensity value in the threat level control vector. Using the spatial coordinates of the biological target at the previous moment as a reference point, the main prediction path is generated by extending the length of the trajectory prediction time window along the direction of motion, and a fan-shaped prediction area is generated using the spatial trajectory expansion radius as the scattering radius. The main predicted path is discretized into a sequence of spatial coordinate points. The azimuth and pitch angles of each coordinate point relative to the unmanned intelligent turntable, as well as the turning angular velocity between adjacent coordinate points, are calculated to generate a spatial angle control signal containing the azimuth angle sequence, pitch angle sequence, and turning angular velocity sequence. The spatial angle control signal is input to the servo controller of the unmanned intelligent turntable. The azimuth and pitch angles of the turntable are continuously adjusted according to the steering angular velocity. When the deviation between the actual direction of the turntable and the target coordinate point exceeds the tolerance angle, the adaptive compensation mechanism is triggered. The compensation angular acceleration is calculated with the actual deviation angle as input and the compensation angular acceleration is superimposed on the steering angular velocity to update the steering parameters.

3. The method according to claim 2, characterized in that, The spatial angle control signal is input to the servo controller of the unmanned intelligent turntable. The azimuth and pitch angles of the turntable are continuously adjusted according to the steering angular velocity. When the actual direction of the turntable deviates from the target coordinate point beyond the tolerance angle, an adaptive compensation mechanism is triggered. The compensation angular acceleration is calculated using the actual deviation angle as input and then superimposed on the steering angular velocity to update the steering parameters, including: The spatial angle control signal is input to the servo controller. The servo controller drives the rotation mechanism of the turntable to continuously adjust the azimuth and pitch angles of the turntable according to the current value in the steering angular velocity sequence, so that the turntable faces the current target position indicated by the azimuth and pitch angle sequences. During the adjustment process, the actual azimuth and actual pitch angles of the turntable are obtained, and the first deviation angle between the actual azimuth and the current value of the azimuth sequence and the second deviation angle between the actual pitch angle and the current value of the pitch sequence are calculated. The magnitudes of the first deviation angle and the second deviation angle are compared with a preset tolerance angle. When either the first deviation angle or the second deviation angle exceeds the tolerance angle, an adaptive compensation mechanism is activated. In the adaptive compensation mechanism, the first deviation angle and the second deviation angle are used as inputs to calculate the first compensation angular acceleration and the second compensation angular acceleration, respectively. The first compensation angular acceleration is superimposed on the azimuth component of the current value of the steering angular velocity sequence to form the updated azimuth steering angular velocity, and the second compensation angular acceleration is superimposed on the pitch component of the current value of the steering angular velocity sequence to form the updated pitch steering angular velocity. The updated azimuth steering angular velocity and pitch steering angular velocity are combined as the overall updated steering parameters. The overall updated steering parameters are used to replace the corresponding values ​​in the original steering angular velocity sequence, and the continuous adjustment process of the turntable direction is continued.

4. The method according to claim 1, characterized in that, The enhanced continuous tracking capability-acquired feature data is input into a pre-constructed spatiotemporal graph convolutional network. This network constructs spatial topology nodes based on biological joint connections, and maps the time-series motion features from the feature data to these spatial topology nodes, forming a motion topology graph. This includes: The feature data acquired under the enhanced continuous tracking capability is transmitted to the data receiving interface of the spatiotemporal graph convolutional network, and a spatial node array is created according to the natural connection structure of biological joints. Each spatial node corresponds to the physical location of a specific biological joint, and connection edges are established between anatomically adjacent nodes. The motion parameter time series of each biological joint in the feature data is assigned to the temporal feature container of the corresponding spatial node; By combining the spatial node array, connecting edges, and loaded temporal feature containers, a motion topology graph that integrates spatial topology and temporal motion features is generated.

5. The method according to claim 1, characterized in that, The motion topology map is used to identify specific abnormal behavior patterns of the biological target within the perimeter area, and these patterns are matched against a perimeter protection threat rule base to generate a threat level control vector, including: Traverse the spatial nodes of the motion topology graph, extract the extreme values ​​and frequencies of the directional angle change sequence from the temporal feature container of each node, calculate the number of sudden accelerations and the acceleration amplitude of the rate change sequence, count the abnormal oscillation period of the amplitude change sequence, and mark the nodes that meet any of the following conditions as abnormal nodes: the directional angle fluctuation frequency exceeds the first set threshold, the number of sudden accelerations exceeds the second set threshold within a unit time, or the abnormal oscillation period is shorter than the third set threshold. An abnormal behavior pattern map is constructed based on the spatial distribution of the abnormal nodes and their connecting edge relationships. The abnormal behavior pattern map is then compared with each rule entry in the perimeter protection threat rule base. When the similarity between the abnormal behavior pattern map and any rule entry exceeds the matching threshold, a threat level control vector containing threat type encoding and threat intensity value is generated.

6. The method according to claim 1, characterized in that, When the detection device integrated into the unmanned intelligent turntable scans the perimeter area, it captures the reflected signals of biological targets and extracts feature data generated by the subtle movements of biological joints from the reflected signals, including: The detection device emits continuous electromagnetic scanning signals to the perimeter area, receives mixed signals reflected by the biological target, and separates the modulation signal component generated by the periodic micro-displacement of the biological joints from the mixed signal. The modulated signal components are divided in the time domain and focused in the frequency domain to extract the oscillation frequency, amplitude fluctuation and phase shift of the signal waveform in each segmented time period; The oscillation frequency is mapped to the joint swing rate, the amplitude fluctuation is mapped to the joint displacement amplitude, and the phase offset is mapped to the joint motion direction angle, generating feature data composed of rate, amplitude, and direction angle.

7. The method according to claim 1, characterized in that, By responding to the beam modulation command through a programmable metasurface, the electromagnetic wave phase distribution of the programmable metasurface is reconstructed, and the direction and coverage of the emitted beam of the detection device are adjusted to enhance the continuous tracking capability of the biological target, including: The target azimuth angle, target elevation angle, and space cone angle range are extracted from the beam control command. Based on the target azimuth and target elevation angles, the phase compensation values ​​required to be generated by each control unit of the programmable metasurface are calculated. The phase compensation values ​​enable the electromagnetic waves emitted by each control unit to form coherent superposition in the direction defined by the target azimuth and target elevation angles. The gradient distribution of the phase compensation values ​​is adjusted according to the spatial cone angle range so that the width of the coherent superposition beam is extended to the spatial cone angle range. The phase compensation value is written into the control unit array of the programmable metasurface, so that the emitted beam forms a dynamic tracking area centered on the target azimuth and elevation angles and covering the angle range of the spatial cone. The phase distribution of the dynamic tracking area is continuously updated as the biological target moves.

8. The method according to claim 1, characterized in that, Based on the location information and movement trajectory of the biological target reflected in the feature data, beam control commands are generated, including: Based on the change in the orientation angle of consecutive timestamps in the feature data, the movement vector of the biological target in three-dimensional space is calculated, and the movement path within the future time window is extrapolated based on the movement vector. The position points on the motion path are converted into spherical coordinates with the unmanned intelligent turntable as the origin, generating a trajectory coordinate sequence containing distance value, azimuth angle, and pitch angle; The range of spatial cone angles that the transmitted beam needs to cover is determined based on the maximum offset distance between adjacent coordinate points in the trajectory coordinate sequence. The azimuth, elevation, and spatial cone angle ranges of the target tracking point are encapsulated as command parameters for beam control commands.

9. An intelligent remote control system for an unmanned intelligent turntable, characterized in that, include: The first generation module captures the reflected signals of biological targets when the detection device integrated into the unmanned intelligent turntable scans the perimeter area, and extracts feature data generated by the subtle movements of biological joints from the reflected signals. The second generation module generates beam control commands based on the orientation information and movement trajectory of the biological target reflected by the feature data. The enhancement module, through the response of the programmable metasurface to the beam control command, reconstructs the electromagnetic wave phase distribution of the programmable metasurface and adjusts the direction and coverage of the emitted beam of the detection device to enhance the continuous tracking capability of the biological target. The forming module inputs the feature data acquired under the enhanced continuous tracking capability into a pre-constructed spatiotemporal graph convolutional network, so as to construct spatial topology nodes based on the biological joint connection relationship through the spatiotemporal graph convolutional network, and map the time series motion features in the feature data to the spatial topology nodes to form a motion topology graph; The third generation module identifies specific abnormal behavior patterns of the biological target within the perimeter area through the motion topology map, and matches the specific abnormal behavior patterns with the perimeter protection threat rule base to generate a threat level control vector. The drive module calculates the trajectory prediction parameters of the biological target based on the threat level control vector, generates a spatial angle control signal for the unmanned intelligent turntable, and drives the unmanned intelligent turntable to perform automatic locking and adaptive tracking actions according to the spatial angle control signal.

10. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the intelligent remote control method for an unmanned intelligent turntable as described in any one of claims 1 to 7.

Images