An unmanned aerial vehicle dynamic space-time regularization tracking device and method for an infrared search and track system

By using a dynamic spatiotemporal regularization method, the spatial regularization weights and memory length are adaptively adjusted. Combined with a memory strategy, the filter update of the infrared search and tracking system is optimized, which solves the drift and occlusion problems of UAV target tracking in the infrared search and tracking system and achieves rapid re-acquisition.

CN122816227BActive Publication Date: 2026-11-10BEIJING ZHONGTIAN XINGKONG TECH DEV CO LTD CHENGDU BRANCH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611230836.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-10
Estimated Expiration
2046-08-14

AI Technical Summary

Technical Problem

In existing infrared search and tracking systems for UAV target tracking, spatial regularization weights cannot adapt to background changes, and temporal regularization single-frame constraints are unreliable, leading to tracking drift and re-capture delays after occlusion.

Method used

A dynamic spatiotemporal regularization method is adopted, which optimizes the filter coefficient update by combining adaptive spatially aware regularization weights and dynamic memory length driven by motion state confidence with memory write gating and occlusion recovery clearing strategies.

Benefits of technology

It effectively alleviates boundary effects and background clutter interference, resolves the contradiction between long memory noise suppression and short memory contamination avoidance, and achieves rapid re-capture after occlusion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122816227B_ABST
    Figure CN122816227B_ABST
Patent Text Reader

Abstract

The application discloses an unmanned aerial vehicle dynamic space-time regularization tracking device and method of an infrared search and tracking system, and belongs to the technical field of unmanned aerial vehicle target tracking. The device comprises the following steps: receiving a first thermal infrared image, calculating a current frame response graph; constructing a space perception regularization weight matrix; calculating a motion state confidence; determining a dynamic memory length according to the motion state confidence, selecting a plurality of recent frame response graphs from a history response stack according to the dynamic memory length to construct a memory reference response; executing a memory write gate and an occlusion recovery emptying strategy; solving filter coefficients and storing the filter coefficients; extracting peak pixel coordinates from the current frame response graph, converting the peak pixel coordinates into azimuth angle deviation and pitch angle deviation, and outputting the azimuth angle deviation and the pitch angle deviation to a servo control unit, while outputting absolute azimuth angle and pitch angle tracks; effectively suppressing background thermal noise interference, using a long history smooth noise when tracking is stable, shortening the memory and cutting off the error history write when the response is unreliable, and reducing the memory inertial drift.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) target tracking technology, and particularly to a dynamic spatiotemporal regularized tracking device and method for UAVs using an infrared tracking system. Background Technology

[0002] An infrared search and track system (IRST) typically consists of an infrared optical system, an infrared focal plane detector (FPA), a stabilization platform or servo mechanism, a signal and tracking processing board, and a display / fire control / alarm / guidance link. The FPA outputs a raw or front-end calibrated sequence of thermal infrared images; after the IRST enters tracking mode, the tracking processing board maintains lock on the UAV target and outputs azimuth / pitch deviation to the servo control unit, driving the optical platform to keep the UAV near the center of the tracking field of view.

[0003] In IRST tracking scenarios, UAV targets typically exhibit weak thermal radiation and occupy few pixels, resulting in blunt peak values ​​and high sidelobes in the correlation filter response. Tracking accuracy is easily affected by background clutter and platform disturbances. In existing technologies, ARCF (Adjacent-frame Response Constraint Filter) is the fundamental model in the field of correlation filter tracking, and its objective function is: ; Among them, the first item This is a data fitting term used to constrain the filter response to approximate a predefined desired response. ; Second item For space regularization, a fixed space weight is used. Apply spatial constraints to the filter coefficients; third term For time regularization, only the response of the previous frame is considered. As a constraint, the response of the current frame is forced to remain consistent with that of the previous frame.

[0004] The model has the following shortcomings: the weights in the space regularization term. The value is fixed and cannot adaptively adjust as the target moves within the field of view. When the target approaches the edge of the field of view or when the thermal clutter distribution changes at the sky-cloud-ground boundary, the filter easily learns background hotspots, leading to tracking drift. The time regularization term only applies to a single frame response. As a constraint, if the response of the previous frame changes abruptly due to UAV fast turning, IRST servo following lag, or occlusion, the error is propagated frame by frame, and it is impossible to use multi-frame historical information to suppress noise.

[0005] Some existing technologies extend the temporal regularization term to a multi-frame weighted fusion form: ; ; Construct a multi-frame temporal memory objective function: ; in For constant memory depth, This represents the objective function value of the existing multi-frame temporal memory model. While this method can suppress random noise during UAV hovering or uniform flight in stable tracking by utilizing multi-frame history, when the UAV enters clouds or is obstructed by buildings, the distortion response is forcibly incorporated into the memory reference, forming memory inertia. This causes the filter to remain dominated by contaminated frames after the obstruction ends, significantly increasing the recapture delay. The root cause lies in: constants There is an irreconcilable contradiction between the two types of operating conditions: long-term memory noise suppression and short-term memory pollution avoidance. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a dynamic spatiotemporal regularized tracking device and method for UAVs in an infrared search and track system. It aims to solve the technical problems in existing technologies where, during continuous tracking of UAV targets in an IRST system, fixed spatial weights cannot adapt to background changes, single-frame constraints are unreliable in the temporal dimension, fixed memory depth is contradictory, and contaminated memory causes inertial drift. This invention effectively alleviates boundary effects and background clutter interference, resolving the contradiction between long-memory noise suppression and short-memory contamination avoidance.

[0007] This invention provides a dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared tracking system, comprising: Receive the The frame thermal infrared image is loaded with the filter coefficients from the previous time step, multi-channel features are extracted, cross-correlation is performed with the corresponding channel filters, and the current frame response map is calculated. The peak position of the current frame response map is obtained as the target center, and a spatially aware regularized weight matrix is ​​constructed. Based on the current frame response map and the previous frame response map, calculate the motion state confidence. The dynamic memory length is determined based on the motion state confidence, and a memory reference response is constructed by selecting several recent frame response maps from the historical response stack based on the dynamic memory length. Execute memory write gating and occlusion recovery clearing strategies based on the motion state confidence level; Substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into a pre-constructed objective function to solve for and store the filter coefficients. The peak pixel coordinates are extracted from the current frame response map, converted into azimuth and pitch deviations, and output to the servo control unit. At the same time, the absolute azimuth and pitch tracks are output.

[0008] In some embodiments, the spatially perceptual regularization weight matrix is ​​obtained by weighted fusion of Gaussian spatial prior and multi-channel local dispersion; the Gaussian spatial prior has the target center as the peak center and increases with increasing distance; the multi-channel local dispersion reflects the degree of deviation of the multi-channel features at each location from the local mean.

[0009] In some embodiments, the motion state confidence is obtained by normalizing and weighting the peak sidelobe ratio, the average peak correlation energy, and the adjacent frame response difference energy; wherein the peak sidelobe ratio is used to characterize the prominence of the response peak, the average peak correlation energy is used to characterize the overall fluctuation of the response map, and the adjacent frame response difference energy is used to characterize the abrupt change of the adjacent frame response map.

[0010] In some embodiments, the dynamic memory length is obtained by mapping the motion state confidence score to a sigmoid function. The higher the motion state confidence score, the longer the dynamic memory length; the lower the motion state confidence score, the shorter the dynamic memory length. The dynamic memory length is limited to a preset lower and upper bound.

[0011] In some embodiments, the memory reference response is constructed by weighted summation of the most recent frame response graphs stored in the historical response stack, with the dynamic memory length as the upper limit of the number of selected frames and exponential decay as the weight; when the actual number of frames stored in the historical response stack is less than the dynamic memory length, the memory reference response is constructed only from the valid frames actually stored in the stack; when the historical response stack is empty, the memory reference response is invalid.

[0012] In some embodiments, the execution of the memory write gating and occlusion recovery clearing strategy based on the motion state confidence includes: When the motion state confidence is lower than the preset freeze threshold, the current frame response map is not written to the historical response stack, the filter coefficient update is paused, and the servo signal is switched to the gate position output by the prediction unit. When the motion state confidence recovers from below the freeze threshold to above the preset recovery threshold, the response records in the occlusion interval in the historical response stack are cleared, the dynamic memory length is reset to the preset lower bound, and the filter coefficients are reinitialized to recover the response peak position of the frame.

[0013] In some embodiments, the prediction unit is a Kalman filter, which performs prediction and update when the motion state confidence is not lower than the freeze threshold, and performs prediction but not update when the motion state confidence is lower than the freeze threshold.

[0014] In some embodiments, the objective function is: ; Among them, the first item This is a data fitting term used to constrain the filter response to approximate a predefined desired response. ; Second item For space regularization, The spatially-aware regularized weight matrix is ​​used to impose spatially differentiated constraints on the filter coefficients; the third term... For time regularization, The memory reference response is used to constrain the filter response to approximate the memory reference response; For the first The filter coefficients to be solved in the frame; This represents the total number of multi-channel feature channels; For the first Frame number Channel feature map; This is the clipping matrix, used to extract sample features from the target center region; This indicates element-wise multiplication. These are the weighting coefficients for the time regularization term; In response to the map location index, it represents the pixel location. The value at that location.

[0015] In some embodiments, the azimuth deviation and pitch deviation are obtained by multiplying the pixel deviation of the peak pixel coordinates relative to the center of the tracking field of view by a preset scaling factor and superimposing a preset zero offset; the absolute azimuth and pitch angles are obtained by superimposing the azimuth deviation and pitch deviations on the current line of sight angle of the stable platform.

[0016] This invention provides a dynamic spatiotemporal regularized tracking device for an infrared search and tracking system for unmanned aerial vehicles (UAVs). Deployed in the tracking processing channel of the infrared search and tracking system after the UAV target has been acquired and locked, the device includes: The FPA data interface module is used to receive thermal infrared image sequences output by the infrared focal plane detector. The feature extraction module is used to extract multi-channel features from the thermal infrared image; The ARCF response calculation module is used to load the filter coefficients from the previous time step, perform cross-correlation operations between the multi-channel features of the current frame and the corresponding channel filters, and calculate the response map of the current frame. The spatially aware regularization module is used to obtain the peak position of the current frame response map as the target center and construct the spatially aware regularization weight matrix. The dynamic memory scheduling module is used to determine the dynamic memory length frame by frame based on the motion state confidence, construct the memory reference response, and execute the memory write gating and occlusion recovery clearing strategy. The ADMM filter solving module is used to substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into a pre-constructed objective function and solve for the filter coefficients. The servo interface module is used to extract peak pixel coordinates from the response map and convert pixel deviations into azimuth and pitch deviations. The first path outputs the azimuth and pitch deviations to the servo control unit of the infrared search and track system to drive the stabilization platform to adjust the line of sight. The second path superimposes the azimuth and pitch deviations with the current line of sight angle of the stabilization platform and outputs them to the track output module of the infrared search and track system to calculate and output the absolute azimuth and pitch angles of the UAV target to the track management and / or alarm / guidance link.

[0017] Compared with the prior art, the present invention has the following beneficial effects: By improving the fixed spatial weights in the spatial regularization term to a frame-by-frame adaptive spatially aware regularization weight matrix, the boundary effect and background clutter interference are effectively mitigated. In the temporal regularization term, the constant memory depth is improved to a dynamic memory length driven by motion state confidence. Combined with memory write gating and occlusion recovery clearing strategies, the memory reference response in the temporal regularization term is always matched with the tracking reliability of the current frame. This simultaneously resolves the contradiction between long memory noise suppression and short memory pollution avoidance, and enables rapid re-capture after occlusion. Attached Figure Description

[0018] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the structure of an infrared tracking system for UAV dynamic spatiotemporal regularization tracking, provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of an IRST provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the operational phases of IRST on a UAV target, as provided in an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the implementation process of an infrared tracking system for UAV dynamic spatiotemporal regularization tracking, provided in an embodiment of the present invention. Figure 5 A schematic diagram of spatially aware regularization weights provided in an embodiment of the present invention; Figure 6 A schematic diagram illustrating the execution memory write gating and occlusion recovery clearing strategy provided in an embodiment of the present invention; Figure 7 This is a comparison curve of constant memory depth and dynamic memory length occlusion recapture provided in an embodiment of the present invention; Figure 8This is a schematic diagram of the composition structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] If the application documents contain similar descriptions such as "first, second, third", the following explanation shall be added: In the following description, the terms "first, second, third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0023] This invention provides an infrared tracking system for UAV dynamic spatiotemporal regularization tracking. The system includes various modules and units, which can be implemented by a processor in a computer device; or by specific logic circuits. In implementation, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0024] This invention provides a dynamic spatiotemporal regularized tracking device for unmanned aerial vehicles (UAVs) using an infrared tracking system. Figure 1 This is a schematic diagram of the structure of an infrared tracking system for UAVs with dynamic spatiotemporal regularization, provided in an embodiment of the present invention. Figure 1As shown, the tracking and processing channel deployed in the infrared search and track system after intercepting and locking onto the UAV target includes: The FPA data interface module is used to receive thermal infrared image sequences output by the infrared focal plane detector. The feature extraction module is used to extract multi-channel features from the thermal infrared image; The ARCF response calculation module is used to load the filter coefficients from the previous time step, perform cross-correlation operations between the multi-channel features of the current frame and the corresponding channel filters, and calculate the response map of the current frame. The spatially aware regularization module is used to obtain the peak position of the current frame response map as the target center and construct the spatially aware regularization weight matrix. The dynamic memory scheduling module is used to determine the dynamic memory length frame by frame based on the motion state confidence, construct the memory reference response, and execute the memory write gating and occlusion recovery clearing strategy. The ADMM filter solving module is used to substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into a pre-constructed objective function and solve for the filter coefficients. The servo interface module is used to extract peak pixel coordinates from the response map and convert pixel deviations into azimuth and pitch deviations. The first path outputs the azimuth and pitch deviations to the servo control unit of the infrared search and track system to drive the stabilization platform to adjust the line of sight. The second path superimposes the azimuth and pitch deviations with the current line of sight angle of the stabilization platform and outputs them to the track output module of the infrared search and track system to calculate and output the absolute azimuth and pitch angles of the UAV target to the track management and / or alarm / guidance link.

[0025] In this embodiment of the invention, the embedded tracking processing device is installed in the form of a signal processing board in the IRST signal processing chassis and deployed in the tracking processing channel after the infrared search and track system has completed the interception and locking of the UAV target. Figure 2 As shown, the IRST system consists of an infrared telescope optical system, an infrared focal plane detector (FPA), an embedded tracking processing unit (signal processing board), a servo control unit, and a stabilization platform. The FPA receives the thermal radiation from the UAV and outputs a sequence of thermal infrared images; the embedded tracking processing unit performs correlation filtering tracking on the intercepted and locked UAV, outputting the azimuth and pitch angle deviations to the servo control unit, which drives the stabilization platform to adjust the line of sight, forming a line-of-sight closed loop; after being superimposed with the platform's current line-of-sight angle by the track output module, the absolute azimuth and pitch angles are output to the track management and / or alarm / guidance link.

[0026] In this embodiment of the invention, the FPA data interface module receives the thermal infrared image sequence output by the infrared focal plane detector via a Camera Link or LVDS interface, with an update rate of 50Hz. The feature extraction module is connected to the FPA data interface module and is used to extract multi-channel features from the thermal infrared images. The ARCF response calculation module loads the filter coefficients from the previous time step, performs cross-correlation operations (using FFT acceleration) on the multi-channel features of the current frame with the corresponding channel filters, and calculates the response map of the current frame. The spatial awareness regularization module obtains the peak position of the response map of the current frame as the target center, constructs a Gaussian spatial prior, calculates the multi-channel local dispersion, and weights and fuses them to generate a spatial awareness regularization weight matrix. This weight matrix has lower values ​​in the neighborhood of the target center and higher values ​​in the sky-ground boundary and background thermal clutter regions, thereby spatially differentiating the filter coefficients. The dynamic memory scheduling module determines the dynamic memory length frame by frame based on the motion state confidence, constructs a memory reference response, and executes memory writing gating and occlusion recovery clearing strategies. The ADMM filter solving module substitutes multi-channel features, spatially aware regularized weight matrix, and memory reference response into a pre-constructed objective function, and uses the alternating direction multiplier method to solve for the optimal filter coefficients. The servo interface module outputs azimuth and pitch deviations to the IRST servo control unit via an RS422 bus at a 50Hz update rate, driving the stabilization platform to adjust its line-of-sight pointing. Simultaneously, it outputs these deviations to the track output module via the board's internal data bus. The track output module receives the azimuth and pitch deviations, reads the current line-of-sight azimuth and pitch angles from the stabilization platform's encoder / INS feedback, and superimposes the angular deviations with the platform's current line-of-sight angles to obtain the absolute azimuth and pitch angles of the UAV target. This information is then output to the track management and alarm / guidance link via Ethernet or serial port.

[0027] To address the problems existing in related technologies, this invention provides a dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared tracking system. The execution entity of this tracking method can be a computer device. The computer device can be a terminal such as a personal computer, laptop, tablet, embedded processor on an IRST signal processing board, vehicle-mounted computing device, or edge computing device, or it can be implemented as a server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0028] In some embodiments, the functions implemented by the tracking method provided in this invention can be achieved by the processor of a computer device calling program code, wherein the program code can be stored in a computer storage medium.

[0029] This invention provides a dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared tracking system. The execution entity of this tracking method can be an embedded tracking processing device. The embedded tracking processing device can be a hardware device installed in the form of a signal processing board within an IRST signal processing chassis, or it can be a software functional module implemented by logic circuits or processor-executed program code integrated on the IRST signal processing board.

[0030] like Figure 3 As shown, IRST divides the working phase of UAV targets into three stages: search and scan, interception and locking, and continuous tracking. This invention accesses and executes the tracking method in the continuous tracking stage.

[0031] This invention provides a dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared tracking system. Figure 4 This is a schematic diagram illustrating the implementation process of a dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared tracking system, as provided in an embodiment of the present invention. Figure 4 As shown, it includes: S1: Receive the first The frame thermal infrared image is loaded with the filter coefficients from the previous time step, multi-channel features are extracted, cross-correlation is performed with the corresponding channel filters, and the current frame response map is calculated. In this embodiment of the invention, the feature extraction module extracts multi-channel features from the current frame of thermal infrared image. The ARCF response calculation module loads the previous frame (the first frame) from the filter buffer. The optimal filter coefficients obtained by solving the frame) The multi-channel features of the current frame are cross-correlated with the corresponding channel filters in the frequency domain (using FFT for acceleration) to calculate the response map of the current frame. .when At that time, the initial filter coefficients are trained based on the image patch at the target location captured in the first frame, and the historical response stack is initialized to empty.

[0032] S2: Obtain the peak position of the current frame response map as the target center, and construct a spatially aware regularized weight matrix; In some embodiments, the spatially perceptual regularization weight matrix is ​​obtained by weighted fusion of Gaussian spatial prior and multi-channel local dispersion; the Gaussian spatial prior has the target center as the peak center and increases with increasing distance; the multi-channel local dispersion reflects the degree of deviation of the multi-channel features at each location from the local mean.

[0033] In this embodiment of the invention, the spatial awareness regularization module obtains the response map of the current frame. The peak position is taken as the target center of the UAV, and a Gaussian space prior is constructed with this target center as the peak center. : ; In the formula, Position of frame t No. The pixel coordinates of the frame UAV target center in the feature map / response map; For the first The broadening parameter of the Gaussian prior controls the rate at which the weights decay from the center to the edge.

[0034] Simultaneously calculate the local dispersion of the multi-channel. : ; In the formula, For the first Frame position The multi-channel local dispersion at a given location reflects the degree of deviation between the location's characteristics and the local mean. For the first Frame position Place Channel characteristics The arithmetic mean; The two are weighted and fused to generate a spatially perceptual regularized weight matrix. : ; In the formula, For the first Frame position Spatial-aware regularization weights at the location; For the Gaussian prior term G and the discreteness term The fusion balance coefficient, Preferably, .

[0035] like Figure 5 As shown, The values ​​are lower in the neighborhood of the target center and higher in the sky-ground boundary and background thermal clutter region, thus spatially differentiating the filter coefficients.

[0036] S3: Calculate the motion state confidence based on the current frame response map and the previous frame response map; In some embodiments, the motion state confidence is obtained by normalizing and weighting the peak sidelobe ratio, the average peak correlation energy, and the adjacent frame response difference energy; wherein the peak sidelobe ratio is used to characterize the prominence of the response peak, the average peak correlation energy is used to characterize the overall fluctuation of the response map, and the adjacent frame response difference energy is used to characterize the abrupt change of the adjacent frame response map.

[0037] In this embodiment of the invention, the dynamic memory scheduling module determines the response based on the current frame. and the response graph of the previous frame Extracting peak sidelobe ratio Average peak correlation energy and the energy difference in response between adjacent frames .in, Used to characterize the prominence of response peaks; a higher value indicates a sharper and more reliable target response, especially during UAV cloud entry / occlusion. It will decrease significantly; Used to characterize the overall fluctuation of the response map, a higher value indicates a more concentrated response map and better distinction between the target and background, especially when a weak target is overwhelmed by clutter. It will decrease; Used to characterize the degree of abrupt changes in the response map between adjacent frames, especially during rapid UAV turns or IRST servo follow lag. It will increase abnormally. The above three parameters reflect the tracking reliability of the current frame from different dimensions. The motion state confidence is obtained by performing min-max normalization on each of the three parameters separately, and then weighted and fused. ; In the formula, represents the confidence level of motion state in frame t; a higher value indicates more reliable UAV tracking. , , Corresponding to The normalized fusion weights, and Preferably, The preferred range is 0.3-0.5. The preferred range is 0.3-0.5. The preferred range is 0.1-0.3, for example, , , ,in, For min-max normalization, Linear mapping to respectively Closed interval: ; In the formula, This refers to the statistical extreme value within the current tracking sequence or sliding window. It is a small constant greater than zero to prevent division by zero.

[0038] S4: Determine the dynamic memory length based on the motion state confidence, and construct a memory reference response by selecting the most recent frame response maps from the historical response stack based on the dynamic memory length; In some embodiments, the dynamic memory length is obtained by mapping the motion state confidence score to a sigmoid function. The higher the motion state confidence score, the longer the dynamic memory length; the lower the motion state confidence score, the shorter the dynamic memory length. The dynamic memory length is limited to a preset lower and upper bound.

[0039] In this embodiment of the invention, the dynamic memory scheduling module will Determining dynamic memory length using Sigmoid mapping : ; In the formula, The length is dynamically memorized frame by frame for the t-th frame; This is a lower bound for the dynamic memory length, when tracking is unreliable. Approaching ; This is the upper bound of the dynamic memory length, when tracking is stable. Approaching ,and Less than or equal to the capacity of the historical response stack; This is a rounding function that guarantees... It is an integer; To be Mapped to The Sigmoid function for an interval.

[0040] The Sigmoid mapping function is: ; In the formula, the Sigmoid slope coefficient, ,control Mapping to the (0,1) interval The response sensitivity; the Sigmoid inflection point position, Mapping equals hour Preferably, , .

[0041] Its function is as follows: When At higher levels (stable tracking segment). Approaching The time regularization term incorporates more historical frame responses, fully leveraging the noise reduction effect of multi-frame averaging to suppress random noise during UAV hovering or constant-speed flight; when At lower levels (cloud entry / obstruction / maneuvering phase). Approaching The number of historical frames incorporated in the time regularization term is reduced, avoiding outdated or distorted responses from contaminating memory references and achieving short-term memory contamination avoidance. Therefore, It can be used within the same tracking sequence depending on the operating conditions. and The frame-by-frame switching resolves the inherent trade-off between the limitations of existing technologies in achieving both long-memory noise suppression and short-memory contamination avoidance under different operating conditions. Preferably, , .

[0042] In some embodiments, the memory reference response is constructed by weighted summation of the most recent frame response graphs stored in the historical response stack, with the dynamic memory length as the upper limit of the number of selected frames and exponential decay as the weight; when the actual number of frames stored in the historical response stack is less than the dynamic memory length, the memory reference response is constructed only from the valid frames actually stored in the stack; when the historical response stack is empty, the memory reference response is invalid.

[0043] In this embodiment of the invention, the dynamic memory scheduling module uses To find the upper limit of the summation, use exponential decay. As weights, the most recent response stored in the historical response stack We construct a memory reference response by performing a weighted summation on the frame response graphs. : ; In the formula, For the first Frame, spatial location Memory reference response at the location; For the first Frame correlation filter response diagram; Among them, exponential decay : ; In the formula, The time decay coefficient, The larger the value, the higher the weight of recent frames and the faster the decay of distant frames; For historical frame backtracking index, Indicates the previous frame. Indicates the previous two frames; For weighted normalized summation index, Preferably, the time decay coefficient .

[0044] When the actual number of frames stored in the historical response stack is less than When the historical response stack is empty, only the valid frames actually stored in the stack are weighted and fused; when the historical response stack is empty, It degenerates into an all-zero matrix.

[0045] S5: Execute the memory write gating and occlusion recovery clearing strategy based on the motion state confidence; In some embodiments, the execution of the memory write gating and occlusion recovery clearing strategy based on the motion state confidence includes: When the motion state confidence is lower than the preset freeze threshold, the current frame response map is not written to the historical response stack, the filter coefficient update is paused, and the servo signal is switched to the gate position output by the prediction unit. When the motion state confidence recovers from below the freeze threshold to above the preset recovery threshold, the response records in the occlusion interval in the historical response stack are cleared, the dynamic memory length is reset to the preset lower bound, and the filter coefficients are reinitialized to recover the response peak position of the frame.

[0046] In this embodiment of the invention, the dynamic memory scheduling module simultaneously executes memory write gating: when At that time, do not display the response graph of the current frame. Write to the historical response stack, pause filter coefficient updates, and switch the servo signal to the gate position output by the prediction unit; when At that time, clear the pollution response records within the occluded area in the historical response stack. Reset to The filter coefficients are reinitialized to recover the peak position of the frame response. Figure 6 As shown, in the stable tracking segment Maintaining a high level, Approaching Historical response stack writes are normal; UAV cloud ingress / obstruction segment, response peak collapse leads to Fall to the following, Approaching The memory is written to the freeze; after recovery rebounded to The above clears the history response stack. Reset to After capturing again, it is rewritten normally. For example... Figure 7 As shown, during the above recovery process, the existing constant memory depth experiences a slow increase in confidence due to the continuous involvement of contaminated frames in memory reference; it takes approximately 18.4 frames to recover to above the recapture threshold. In contrast, the dynamic memory depth of this invention… The memory was shortened and the write was frozen during the occluded segment. There were no contaminated frames in the memory stack during recovery, and recapture was completed in about 11.2 frames, which is a relative reduction of about 39.1%.

[0047] In some embodiments, the prediction unit is a Kalman filter, which performs prediction and update when the motion state confidence is not lower than the freeze threshold, and performs prediction but not update when the motion state confidence is lower than the freeze threshold.

[0048] In this embodiment of the invention, when When, the prediction unit performs prediction and update; when At this time, the prediction unit only performs prediction and does not perform updates. Preferably, Pick 0.6 times the moving average, Pick 0.8 times the moving average.

[0049] S6: Substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into the pre-constructed objective function to solve for and store the filter coefficients; In some embodiments, the objective function is: ; Among them, the first item This is a data fitting term used to constrain the filter response to approximate a predefined desired response. ; Second item For space regularization, The spatially-aware regularized weight matrix is ​​used to impose spatially differentiated constraints on the filter coefficients; the third term... For time regularization, The memory reference response is used to constrain the filter response to approximate the memory reference response; For the first The filter coefficients to be solved in the frame; This represents the total number of multi-channel feature channels; For the first Frame number Channel feature map; This is the clipping matrix, used to extract sample features from the target center region; This indicates element-wise multiplication. These are the weighting coefficients for the time regularization term; In response to the map location index, it represents the pixel location. The value at that location. Preferably, .

[0050] In this embodiment of the invention, the ADMM filter solving module will solve for multi-channel features. Spatial-aware regularized weight matrix Memory Reference Response Substituting the pre-constructed objective function, the optimal filter coefficients are solved using the alternating direction multiplier method (ADMM). The ADMM solution process includes: introducing auxiliary variables, constructing an augmented Lagrangian function, alternately updating the variables, and terminating the iteration early when the principal residual and dual residual are less than a preset tolerance. The obtained optimal filter coefficients are then... Stored in the filter buffer for use in the next frame.

[0051] S7: Extract the peak pixel coordinates from the current frame response map, convert them into azimuth and pitch deviations and output them to the servo control unit, while outputting the absolute azimuth and pitch tracks.

[0052] In some embodiments, the azimuth deviation and pitch deviation are obtained by multiplying the pixel deviation of the peak pixel coordinates relative to the center of the tracking field of view by a preset scaling factor and superimposing a preset zero offset; the absolute azimuth and pitch angles are obtained by superimposing the azimuth deviation and pitch deviations on the current line of sight angle of the stable platform.

[0053] In this embodiment of the invention, the servo interface module receives the response graph from the current frame. Peak pixel coordinates are extracted and converted into azimuth and pitch deviations using pixel-angle deviation conversion. These deviation signals are output in two paths: the first path is output to the IRST servo control unit via an RS422 bus to drive the stabilization platform to adjust its line-of-sight (LOS) pointing. The second path is output to the track output module via the board's internal data bus. The track output module reads the current LOS azimuth and pitch angles from the stabilization platform's encoder / INS feedback, and superimposes these deviations onto the platform's current LOS angle to obtain the absolute azimuth and pitch angles. These are then output to the track management and / or alarm / guidance link via Ethernet or serial port. The specific steps are as follows: Response diagram Peak pixel coordinate extraction: Response peak location: ; In the formula, For the first Frame current frame response graph The column and row coordinates of the pixel containing the global peak value; To track the search area, the center is usually located at... Candidate window; To make the current frame response graph The pixel position operator that takes the maximum value.

[0054] Pixel deviation relative to the center of the tracking field of view: Pixel deviation: ; In the formula, For the first Frame response peak relative to the center of the tracking field of view Horizontal and vertical pixel deviation; For the first Frame tracking field of view center pixel coordinates; during normal tracking, take the peak value of the previous frame or the image center. In some cases, the gate center output by the prediction unit (such as Kalman filter) can be used instead.

[0055] Pixel deviation to servo angle deviation conversion (IRST installation calibration): The servo control unit receives the azimuth deviation. pitch angle deviation (Unit: degrees or mrad), not pixel coordinates. The conversion is performed based on the installation calibration parameters of the IRST optical-geometry link: Azimuth / Pitch Angle Deviation: ; ; In the formula, , This refers to the pixel-angle deviation scaling factor for azimuth and pitch directions, in units of: Or mrad / pixel, obtained from IRST factory or field calibration; , It is the zero-position offset compensation amount in the azimuth and pitch directions, used to correct the assembly error between the optical axis and the FPA optical axis.

[0056] Among them, the calibration coefficients are related to the horizontal / vertical instantaneous field of view angles of the IRST tracking sub-field of view. , and the number of pixels in the corresponding direction , The relationship is: .

[0057] Calibration instructions are given using a fixed anti-drone IRST equipment as an example: Let the tracking sub-field of view be defined. Response search area On-site calibration If the first Frame response peak coordinates Tracking Center Then pixel deviation , ; proportionality coefficient The azimuth deviation is obtained after conversion. Pitch angle deviation The servo control unit uses this information to drive the platform and correct the peak direction of the line of sight. The track output module reads the platform encoder. , .

[0058] Absolute bearing / pitch angle output of flight track: Absolute bearing / pitch angle (track): ; ; In the formula, , For the first The absolute azimuth and elevation angles of the UAV target under the IRST station's centroid are sent by the track output module to the track management and / or alarm / guidance link. , For the first The current azimuth and pitch angles of the frame IRST stabilized platform are read by the platform encoder or inertial navigation interface.

[0059] Servo interface module with , Forming closed-loop control input (update rate, e.g., 50Hz); the track output module uses... , They form absolute directional quantities, which have the same origin but different uses.

[0060] Comparative verification: The present invention will be dynamically applied to the IRST anti-drone correlation sequences in the LSOTB-TIR dataset. plan( , ) and existing technology constant memory depth schemes ( Using formula A comparative evaluation was conducted using constructed memory references, and the results are shown in Table 1: Table 1 The above results indicate that the present invention is dynamic The proposed solution exhibits significant advantages over existing technologies with constant memory depth in terms of occlusion-related attributes and recapture time, improving the success rate of occlusion attributes by 9.0% and reducing the number of recapture frames by 39.1%. The fundamental reason is that constant memory depth still forcibly incorporates distortion responses into the memory reference during UAV cloud entry / occlusion, creating memory inertia; while this invention has already passed [a certain standard] in the occlusion segment. Will shortened to It freezes memory writes and clears the memory stack upon recovery, structurally severing the contaminated memory chain, thus enabling faster recapture.

[0061] Parameter configuration for typical application scenarios: The tracking method and apparatus described above are adaptable to various IRST equipment platforms. While maintaining the core tracking algorithm, only a few configurable parameters need to be adjusted to address the differences in operating conditions across different application scenarios, as follows: I. Fixed Anti-Drone IRST Scenario 1. System Deployment The embedded tracking processing unit is installed in a fixed anti-UAV IRST signal processing chassis in the form of a PMC / VPX signal processing board. The FPA data interface module receives the thermal infrared image sequence output by the mid-wave infrared FPA through a Camera Link or LVDS interface, with an update rate of 50Hz. After target acquisition and locking, the device performs dynamic spatiotemporal regularized tracking processing according to steps S1 to S7.

[0062] 2. Servo output calibration The servo interface module outputs azimuth and pitch deviations to the IRST servo control unit via an RS422 bus, with a refresh rate of 50Hz. The pixel-to-angle deviation conversion method is described in step S7; in this scenario, the calibration parameters are consistent with those in the example of step S7.

[0063] The track output module reads the current line-of-sight azimuth and pitch angles fed back by the platform encoder, obtains the absolute azimuth and pitch angles according to the superposition formula in step S7, and outputs them to the track management and / or alarm / guidance link via Ethernet or serial port.

[0064] 3. Parameter Configuration Fixed IRSTs are typically deployed at fixed ground positions, where the line of sight is less affected by platform disturbances. The following configuration is recommended: , , , , , , , .

[0065] Typical operating condition response: UAV cloud access Falling below the freeze threshold Approaching The memory is written to the freeze, and the servo signal switches to the gate position predicted by the Kalman filter; after emerging from the clouds... Once the condition recovers above the recovery threshold, clear the contaminated records within the occluded region from the historical response stack. Reset to This enables rapid recapture.

[0066] II. Shipborne IRST tracking scenario of low-altitude UAV over the sea surface 1. System Deployment The embedded tracking processing unit is installed in the ship's signal processing chassis and connected to the bridge infrared search and track equipment. The shipborne IRST continuously tracks small UAVs approaching the ship amidst sea clutter and sky background. Its hardware composition and core algorithm are consistent with those of the fixed scenario, the difference being in parameter configuration and downstream interface.

[0067] 2. Differences in parameter configuration Shipborne IRSTs face line-of-sight disturbances caused by hull rolling, manifested as: An abnormal increase, but not a target maneuver. To differentiate between "platform disturbance" and "target loss," the following differentiated configuration is recommended: , , , , , , , .

[0068] The above weights can be fine-tuned according to actual sea conditions during factory calibration or on-site testing. During stable navigation... Approaching Suppressing thermal clutter noise from the sea surface; during high winds and waves Automatic To approximate and avoid false responses introduced by disturbances in the learning platform.

[0069] 3. Differences in output interfaces The shipborne IRST also outputs azimuth and elevation deviations to the fire control channel of the shipborne weapon system via a servo interface module, guiding naval guns, close-in weapon systems, or electronic jamming equipment to intercept targets. Track output is sent to the shipborne track management and situational awareness system using the same formula.

[0070] It should be noted that, in the embodiments of the present invention, if the above-described tracking method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of the present invention are not limited to any specific hardware and software combination.

[0071] Accordingly, embodiments of the present invention provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the tracing method provided in the above embodiments.

[0072] This embodiment provides a computer device, such as... Figure 8 As shown, the computer device includes a processor, a memory, an input / output interface, a communication interface, an input device, and a display unit. The memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above-described method embodiments.

[0073] It should be noted that the descriptions of the storage media and computer device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage media and device embodiments of the present invention, please refer to the descriptions of the method embodiments of the present invention for understanding.

[0074] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of the invention, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the above-described embodiments of the invention are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0075] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, object, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, object, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, object, or apparatus that includes that element.

[0076] In the several embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0077] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0078] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0079] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0080] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a controller to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0081] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic spatiotemporal regularized tracking method for unmanned aerial vehicles (UAVs) using an infrared search and tracking system, characterized in that, include: Receive the The frame thermal infrared image is loaded with the filter coefficients from the previous time step, multi-channel features are extracted, cross-correlation is performed with the corresponding channel filters, and the current frame response map is calculated. The peak position of the current frame response map is obtained as the target center, and a spatially aware regularized weight matrix is ​​constructed. Based on the current frame response map and the previous frame response map, the motion state confidence is calculated; the motion state confidence is obtained by weighted fusion of peak sidelobe ratio, average peak correlation energy and adjacent frame response difference energy after normalization. The peak-to-sidelobe ratio is used to characterize the prominence of the response peak, the average peak correlation energy is used to characterize the overall fluctuation of the response map, and the adjacent frame response difference energy is used to characterize the abrupt change of the adjacent frame response map. The dynamic memory length is determined based on the motion state confidence level. A memory reference response is constructed by selecting several recent frame response maps from the historical response stack based on the dynamic memory length. The dynamic memory length is obtained by mapping the motion state confidence level using a Sigmoid function. Higher motion state confidence levels result in a longer dynamic memory length, while lower motion state confidence levels result in a shorter dynamic memory length. The dynamic memory length is limited to a preset lower and upper bound. The memory reference response is constructed by weighted summing of several recent frame response maps stored in the historical response stack, using the dynamic memory length as the upper limit for the number of selected frames and a weight that decays exponentially with the increase of the historical frame backtracking index. When the actual number of frames stored in the historical response stack is less than the dynamic memory length, the memory reference response is constructed only from the valid frames actually stored in the stack. When the historical response stack is empty, the memory reference response is invalid. Execute memory write gating and occlusion recovery clearing strategies based on the motion state confidence level; The step of executing the memory write gating and occlusion recovery clearing strategy based on the motion state confidence includes: When the motion state confidence is lower than the preset freeze threshold, the current frame response map is not written to the historical response stack, the filter coefficient update is paused, and the servo signal is switched to the gate position output by the prediction unit; when the motion state confidence recovers from below the freeze threshold to above the preset recovery threshold, the response records in the occlusion interval in the historical response stack are cleared, the dynamic memory length is reset to the preset lower bound, and the filter coefficients are reinitialized to recover the response peak position of the frame. Substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into a pre-constructed objective function to solve for and store the filter coefficients. The peak pixel coordinates are extracted from the current frame response map, converted into azimuth and pitch deviations, and output to the servo control unit. At the same time, the absolute azimuth and pitch tracks are output.

2. The method according to claim 1, characterized in that, The spatial perception regularization weight matrix is ​​obtained by weighted fusion of Gaussian spatial prior and multi-channel local dispersion; the Gaussian spatial prior has the target center as the peak center and increases with increasing distance; the multi-channel local dispersion reflects the degree of deviation of the multi-channel features at each location from the local mean; wherein: Get the current frame response graph The peak position is taken as the target center of the UAV, and a Gaussian space prior is constructed with this target center as the peak center. : ; In the formula, Position of frame t Prior weights in the Gaussian space at the location; No. The pixel coordinates of the frame UAV target center in the feature map / response map; For the first The broadening parameter of the Gaussian prior.

3. The method according to claim 1, characterized in that, The prediction unit is a Kalman filter. When the confidence level of the motion state is not lower than the freezing threshold, prediction and update are performed. When the confidence level of the motion state is lower than the freezing threshold, only prediction is performed and no update is performed.

4. The method according to claim 1, characterized in that, The objective function is: ; Among them, the first item This is a data fitting term used to constrain the filter response to approximate a predefined desired response. ; Second item For space regularization, The spatially-aware regularized weight matrix is ​​used to impose spatially differentiated constraints on the filter coefficients; the third term... For time regularization, The memory reference response is used to constrain the filter response to approximate the memory reference response; For the first The filter coefficients to be solved in the frame; This represents the total number of multi-channel feature channels; For the first Frame number Channel feature map; This is the clipping matrix, used to extract sample features from the target center region; This represents element-wise multiplication. These are the weighting coefficients for the time regularization term; In response to the map location index, it represents the pixel location. The value at that location.

5. The method according to claim 1, characterized in that, The azimuth deviation and pitch deviation are obtained by multiplying the pixel deviation of the peak pixel coordinates relative to the center of the tracking field of view by a preset scaling factor and adding a preset zero offset; the absolute azimuth and pitch angles are obtained by adding the azimuth deviation and pitch deviation to the current line of sight angle of the stable platform.

6. A dynamic spatiotemporal regularized tracking device for unmanned aerial vehicles (UAVs) using an infrared search and tracking system, characterized in that, Deployed in the tracking and processing channel after the infrared search and track system has acquired and locked onto the UAV target, it includes: The FPA data interface module is used to receive thermal infrared image sequences output by the infrared focal plane detector. The feature extraction module is used to extract multi-channel features from the thermal infrared image; The ARCF response calculation module is used to load the filter coefficients from the previous time step, perform cross-correlation operations between the multi-channel features of the current frame and the corresponding channel filters, and calculate the response map of the current frame. The spatially aware regularization module is used to obtain the peak position of the current frame response map as the target center and construct the spatially aware regularization weight matrix. A dynamic memory scheduling module is used to determine the dynamic memory length frame by frame based on motion state confidence, construct a memory reference response, and execute memory write gating and occlusion recovery clearing strategies. The motion state confidence is obtained by weighted fusion of peak-to-sidelobe ratio, average peak correlation energy, and adjacent frame response difference energy after normalization. The peak-to-sidelobe ratio characterizes the prominence of the response peak, the average peak correlation energy characterizes the overall fluctuation of the response map, and the adjacent frame response difference energy characterizes the abrupt change in the response map of adjacent frames. The dynamic memory length is obtained by mapping the motion state confidence using a Sigmoid function; a higher motion state confidence results in a longer dynamic memory length, and a lower motion state confidence results in a shorter dynamic memory length. The dynamic memory length is limited to a preset lower and upper bound. The memory reference response uses the dynamic memory length as the selected frame number. The upper limit is constructed by weighting and summing the response maps of the most recent frames stored in the historical response stack with a weight that decays exponentially with the increase of the historical frame backtracking index. When the actual number of frames stored in the historical response stack is less than the dynamic memory length, the memory reference response is constructed only from the valid frames actually stored in the stack. When the historical response stack is empty, the memory reference response is invalid. The memory write gating and occlusion recovery clearing strategy is executed according to the motion state confidence, including: when the motion state confidence is lower than the preset freeze threshold, the current frame response map is not written to the historical response stack, the filter coefficient update is paused, and the servo signal is switched to the gate position output by the prediction unit; when the motion state confidence recovers from below the freeze threshold to above the preset recovery threshold, the response records in the occlusion interval in the historical response stack are cleared, the dynamic memory length is reset to the preset lower bound, and the filter coefficients are reinitialized at the response peak position of the recovered frame. The ADMM filter solving module is used to substitute the multi-channel features, the spatially aware regularized weight matrix, and the memory reference response into a pre-constructed objective function and solve for the filter coefficients. The servo interface module is used to extract peak pixel coordinates from the response map and convert pixel deviations into azimuth and pitch deviations. The first path outputs the azimuth and pitch deviations to the servo control unit of the infrared search and track system to drive the stabilization platform to adjust the line of sight. The second path superimposes the azimuth and pitch deviations with the current line of sight angle of the stabilization platform and outputs them to the track output module of the infrared search and track system to calculate and output the absolute azimuth and pitch angles of the UAV target to the track management and / or alarm / guidance link.