A tracking and control method for an optoelectronic ball UAV based on a multi-stage state machine

Abstract

This invention relates to the field of UAV countermeasures technology, and in particular to a multi-stage state machine-based electro-optical ball UAV tracking and control method. The core of this method is the establishment of a three-stage state machine encompassing initial detection, target detection, and tracking. Precise tracking is achieved through radar-guided coarse positioning, electro-optical information confirmation, and adaptive zoom control. This method solves the problems of unstable target tracking, untimely response, and false alarm interference caused by the single control strategy in traditional radar-electro-optical fusion systems. This invention significantly improves the target acquisition success rate and tracking stability of UAVs in complex environments by employing differentiated turning thresholds, multi-counter fusion decision-making, and a dual-mode zoom strategy, thereby enhancing the system's anti-interference capability and overall control efficiency.

Description

Technical Field

[0001] This invention relates to the field of drone countermeasures technology, specifically to a method for tracking and controlling an electro-optical ball drone based on a multi-stage state machine. Background Technology

[0002] As an emerging threat in the non-traditional security field, drones pose a serious challenge to traditional defense systems due to their low-altitude, slow-speed, and small-target characteristics. To effectively counter such targets, multi-source fusion detection and countermeasure systems are gradually becoming the core means of drone defense systems. Among them, radar-electro-optical fusion systems, with their composite detection advantages, are widely used in scenarios such as key area protection, border monitoring, and support for important events.

[0003] Each sensor in this fusion system has inherent limitations. The radar has limited angular resolution, achieving only meter-level positioning accuracy for long-range small UAVs. There are issues with false alarms and missed alarms; false targets mislead the optoelectronic equipment's search, and missed alarms result in overlooked threats. Furthermore, it cannot provide detailed target characteristics, making threat assessment difficult.

[0004] Optoelectronic equipment faces a trade-off between field of view and detection accuracy. A wide field of view is advantageous for searching but lacks accuracy, while a narrow field of view offers high accuracy but limits range. Highly maneuvering targets can easily lead to tracking loss. Fixed zoom parameters cannot adapt to changes in target distance and detection phase, resulting in unstable tracking performance.

[0005] Existing fusion control methods have significant drawbacks. Simple data-level fusion directly converts radar position into photoelectric pointing angle, ignoring differences in sensor time synchronization and accuracy matching. Fixed threshold control strategies cannot meet the requirements of the detection phase: they are too conservative during initial detection and sluggish during precise tracking. After target loss, there is a lack of effective re-acquisition mechanisms, requiring a new global search, resulting in low response efficiency.

[0006] The fundamental problem with traditional control methods lies in the lack of differentiated strategies for different detection stages. From target detection to stable tracking, the process involves stages such as coarse localization, target confirmation, and precise tracking, each with different requirements for control accuracy and response speed. The coarse localization stage requires rapid response to radar guidance, the target confirmation stage requires accumulating verification information, and the precise tracking stage requires high-precision real-time following. Existing methods use uniform control parameters, failing to address the specific needs of each stage.

[0007] A multi-dimensional fusion decision mechanism is lacking. Single decision conditions are susceptible to transient disturbances, leading to false state jumps and tracking instability. Multiple independent verification mechanisms need to be established, and cross-validation should be used to improve decision reliability. Summary of the Invention

[0008] The purpose of this invention is to provide a tracking and control method for an electro-optical ball unmanned aerial vehicle (UAV) based on a multi-stage state machine. This method automatically switches control strategies according to the target detection process: a relaxed following strategy is used for rapid target localization in the initial detection stage; verification information is accumulated in the target detection stage to ensure target authenticity; and an aggressive strategy is used in the tracking stage to maintain stable locking. Combined with adaptive zoom and multi-counter fusion decision-making, intelligent control of the electro-optical equipment is achieved. By constructing a multi-stage state machine and an adaptive zoom algorithm, a differentiated control strategy is provided, which can quickly respond to radar target guidance while ensuring tracking stability and accuracy. This meets the high-performance control requirements of anti-UAV systems for electro-optical equipment and solves the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A tracking and control method for an electro-optical ball unmanned aerial vehicle (UAV) based on a multi-stage state machine includes the following steps:

[0011] Step 1, Radar-guided coarse positioning: Receive radar target information, and calculate the target pointing angle of the photoelectric ball through coordinate system transformation and error compensation;

[0012] Step 2, photoelectric confirmation and status judgment: Establish a three-stage state machine including initial detection, target detection and tracking status. Based on the matching results of radar and photoelectric data, combined with the multi-counter fusion decision mechanism, realize dynamic switching between states;

[0013] Step 3, Adaptive zoom control: Based on the target distance, radar positioning error, and preset pixel requirements, a dual-mode zoom strategy is adopted to dynamically adjust the field of view of the optoelectronic device;

[0014] Step 4, Precise Tracking Control: In tracking mode, an aggressive steering threshold strategy based on angle deviation is adopted, where the aggressive steering threshold θ is... threshold It is a dynamic value, and combines target motion prediction and tracking quality assessment Q. track This enables high-precision and stable tracking.

[0015] Preferably, for the three-stage state machine described in step two, the switching logic is as follows:

[0016] In the initial detection state, a lenient steering threshold is used for rapid response radar guidance;

[0017] Once the target in the initial detection state has been successfully verified by the spatial and temporal consistency of radar and photoelectric data, the system switches to target detection state and pauses turning and zooming operations to accumulate verification information.

[0018] When the consecutive success counter in the target detection state reaches the first preset threshold, the system switches to the tracking state.

[0019] In tracking mode, an aggressive turning threshold is used to maintain target lock. If the number of consecutive lost frames exceeds the second preset threshold, the system will revert to the initial detection state.

[0020] Preferably, regarding step two, the multi-counter fusion decision mechanism includes:

[0021] The photoelectric counter is used to record the number of consecutive frames in which no photoelectric target is detected. When its value exceeds the third preset threshold, an alarm or state rollback is triggered.

[0022] A consecutive success counter is used to record the number of frames that have been successfully matched with electro-optical and radar targets, serving as the main decision condition for transitioning from the target detection state to the tracking state.

[0023] The stable tracking counter is used to record the number of stable frames in which the target is in the center of the field of view during tracking, which serves as the condition for triggering telephoto mode switching.

[0024] Preferably, regarding step three, the dual-mode zoom strategy includes:

[0025] In the short focal length mode, the field of view is calculated based on the coverage radius and target distance synthesized from the radar positioning error, target maneuvering error, and altitude error, which is used to ensure coverage of the target uncertainty area during the coarse positioning stage.

[0026] The telephoto mode calculates the focal length based on the focal length required to achieve a preset optimal number of pixels for the target in the image, and is used to provide high-precision imaging during the tracking phase.

[0027] Preferably, the field of view of the short focal length mode is calculated using the following formula:

[0028]

[0029] Among them, R cover The coverage radius is obtained by taking the square root of the sum of the squares of the position error, maneuver error, and altitude error, and multiplying it by a safety factor; d is the target slant range.

[0030] Preferably, for step one, the coordinate system transformation includes:

[0031] Convert the target geographic coordinate system (λ, φ, h) provided by the radar into a geocentric rectangular coordinate system (X, Y, Z);

[0032] Transform the geocentric rectangular coordinate system into a station-centric northeast-sky coordinate system (E, N, U) with the observation station as the origin.

[0033] The azimuth and elevation angles required for the photoelectric sphere are calculated based on the station-centered coordinate system, and atmospheric refraction corrections are added.

[0034] Preferably, for step one, the error compensation uses a linear model that includes installation deviation, temperature drift correction and gravity deformation correction to compensate for the calculated azimuth and pitch angles, so as to obtain the final control command angle.

[0035] Preferably, for step four, the aggressive steering threshold θ threshold Calculated using the following formula:

[0036]

[0037] in, Typically, the FOV is taken as 0.05-0.15. current This represents the current instantaneous field of view.

[0038] Preferably, for step four, the target motion prediction uses a linear extrapolation method, which predicts the target's possible position during the loss period based on the target's last known position and average velocity vector before loss, for target re-acquisition.

[0039] Preferably, regarding step four, the quality assessment Q is tracked. track Confidence level C is identified through weighted calculation. conf Stability score S stable The position score is obtained by subtracting the ratio of the total deviation of the target in the field of view to the current field of view angle from 1. This evaluation result is used to trigger adaptive adjustment of control parameters or switching of tracking strategy.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. This invention, through a multi-stage state machine-based electro-optical ball UAV tracking and control method, establishes a three-stage state machine and a multi-counter fusion decision mechanism, realizing intelligent progressive control from radar coarse positioning to electro-optical precise tracking. This significantly improves the target acquisition success rate and tracking stability of the anti-UAV system, and is particularly suitable for long-distance small target detection scenarios in complex environments.

[0042] 2. This invention achieves a dynamic balance between the field of view and detection accuracy of optoelectronic devices through a dual-mode adaptive zoom strategy, combining a distance-based short-focus coverage algorithm and a pixel-optimized long-focus algorithm. This effectively solves the technical problem that traditional fixed-focus systems cannot balance search range and recognition accuracy, providing an efficient optoelectronic control solution for anti-drone systems.

[0043] 3. This invention establishes a reliable state transition decision system through multi-counter cross-validation and timing control mechanism, avoiding the problem that a single decision condition is easily affected by instantaneous interference, improving the robustness and anti-interference ability of the system in complex electromagnetic environment and under high target maneuvering conditions, and significantly reducing the error rate and system maintenance cost.

[0044] 4. This invention uses a differentiated steering threshold control strategy to adaptively adjust the response sensitivity of the photoelectric ball according to the detection stage. This ensures both a rapid response to radar guidance in the initial detection stage and precise locking in the tracking stage, effectively improving the response efficiency and control performance of the entire anti-drone system. Attached Figure Description

[0045] Figure 1 This is a flowchart of the detection method of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] To address the issues of simplistic control strategies and unstable target tracking in traditional radar-optoelectronic fusion systems, please refer to [link / reference]. Figure 1 This embodiment provides the following technical solution:

[0048] The multi-stage state machine and adaptive zoom algorithm of the present invention include the following steps:

[0049] Step 1: Radar-guided coarse positioning

[0050] The radar-guided coarse positioning step aims to establish a unified processing framework for multi-source sensor data. Through precise coordinate system transformation, it effectively converts radar spatial information into control commands for optoelectronic devices, providing a reliable initial positioning foundation for subsequent precise tracking. This step receives target spatial location information provided by the radar system as input, including the target's geographic coordinates, altitude information, and data quality identifiers.

[0051] 1.1 Standardized Fusion Processing of Multi-Source Radar Data

[0052] To address the diverse radar data sources potentially present in anti-drone systems, a unified data format conversion and quality assessment mechanism should be established. Due to differences in technical systems and data formats, different radar systems exhibit inconsistencies in target information output regarding field definitions, accuracy levels, and time bases. By designing a standardized data interface layer, location information in various formats can be uniformly converted into a standard data structure within the system. Simultaneously, quality identification information from the data sources is extracted and retained, providing a reliable basis for subsequent fusion and decision-making.

[0053] 1.2 Conversion from Geographic Coordinate System to Geocentric Coordinate System

[0054] A precise transformation model was established from the geographic coordinate system (longitude λ, latitude φ, altitude h) to the geocentric rectangular coordinate system (ECEF). The geocentric coordinate system transformation is fundamental to the entire angle calculation, and its accuracy directly affects the final photoelectric pointing accuracy. The transformation formula is based on the WGS84 ellipsoid parameters:

[0055]

[0056] Where X, Y, and Z are the coordinates of the target in the geocentric rectangular coordinate system, in meters.

[0057] λ, φ, h: the longitude, latitude, and altitude of the target. λ and φ are in radians, and h is in meters.

[0058] N is the radius of curvature of the prime meridian, in meters. It represents the radius from the normal direction at a point on the ellipsoid to the axis of rotation, and is a function of latitude φ. The calculation formula is:

[0059]

[0060] Parameter 'a' is the semi-major axis of the ellipsoid, typically taken as 6378137 meters, and 'e' is the first eccentricity, calculated as follows:

[0061]

[0062] Where b is the minor semi-axis of the ellipsoid, usually taken as 6356752.314245 meters.

[0063] This transformation method takes into account the true shape of the Earth's ellipsoid. Compared with the spherical approximation model, it can significantly reduce coordinate transformation errors in high-precision applications, which is of great significance, especially in the processing of high-latitude regions and high-altitude targets.

[0064] 1.3 Transformation from Geocentric Coordinate System to Station-Centered Coordinate System

[0065] A rotational transformation model is established from the geocentric rectangular coordinate system to the station-centric northeast-sky coordinate system (ENU). The station-centric coordinate system has the observation station as the origin, with the eastward direction as the X-axis, the northward direction as the Y-axis, and the celestial direction as the Z-axis, providing an ideal coordinate framework for calculating azimuth and elevation angles. The transformation is achieved through a three-dimensional rotation matrix:

[0066]

[0067] Wherein, [ENU] is a three-dimensional column vector representing the coordinates of the target point in the Northeast-Eastern Sky (ENU) coordinate system after transformation, in meters;

[0068] E (East): Eastward component, the distance of the target point relative to the observer in the due east direction;

[0069] N (North): North component, the distance of the target point relative to the observer's due north direction;

[0070] U (Up): The vertical (upward) component, the distance of the target point relative to the observer's vertical height (upward).

[0071] The rotation matrix R is constructed based on the geographical location of the observation station:

[0072]

[0073] in and These are the longitude and latitude of the observation station, respectively. This conversion method transforms the target location in the global coordinate system into a local coordinate system centered on the observation station, which facilitates subsequent angle calculations and geometric analysis.

[0074] P target This represents the coordinates of the target point in the geocentric rectangular coordinate system (ECEF), and its form is usually [X...]. target Y target Z target ]^T, the unit is meters;

[0075] P observer This represents the coordinates of the observer (or station center) in the geocentric rectangular coordinate system (ECEF), typically in the form [X...]. observer Y observer Z observer ]^T, the unit is meters;

[0076] 1.4 Calculation of photoelectric pointing angle and geometric projection

[0077] Based on the target position coordinates in the station-centered coordinate system, the required azimuth angle Az and elevation angle El are calculated for the optoelectronic equipment through spherical coordinate transformation. The azimuth angle calculation uses a four-quadrant arctangent function to ensure the correct quadrant for the angle.

[0078]

[0079] Where Az is the azimuth angle, which is the horizontal angle between the target direction line and the direction of the target point, starting from due north and rotating clockwise; E is the east component, which is the displacement of the target point in the east-west direction relative to the observer or the origin of the coordinate system; and N is the north component, which is the displacement of the target point in the north-south direction relative to the observer or the origin of the coordinate system.

[0080] Pitch angle calculation is based on the target's celestial component and horizontal range:

[0081]

[0082] Where El is the elevation angle, which is the vertical angle from the horizontal plane to the target direction line, in degrees;

[0083] U is the displacement or coordinate value of the target point relative to the observer or the origin in the vertical direction;

[0084] E represents the displacement or coordinate value of the target point relative to the observer or the origin in the east-west direction; N represents the displacement or coordinate value of the target point relative to the observer or the origin in the north-south direction.

[0085] Angle calculations require consideration of the periodicity and continuity of angles. The azimuth angle ranges from -π to π, and needs to be converted to a suitable angle range based on the mechanical constraints of the optoelectronic equipment. The elevation angle is typically limited to the range of -π / 2 to π / 2, corresponding to the physical elevation angle limitations of the optoelectronic equipment. To improve the accuracy of angle calculations, atmospheric refraction corrections also need to be considered.

[0086] Atmospheric refraction angle The calculations are based on the target pitch angle and an atmospheric model:

[0087]

[0088] in The atmospheric refractive index is typically between 0.13 and 0.20, with the specific value determined based on atmospheric conditions.

[0089] 1.5 System Error Compensation and Dynamic Calibration

[0090] A compensation model for systematic deviations between radar and optoelectronic equipment is established. The systematic error mainly originates from equipment installation deviations. Machining errors and environmental factors The compensation model employs a linear deviation correction method:

[0091]

[0092]

[0093] Among them, Az corrected This is the corrected azimuth angle, in degrees. and For fixed deviation compensation amount, and This is the temperature correction factor, where T is the current temperature. For reference temperature, For the platform tilt angle, This is the correction factor for gravity deformation.

[0094] 1.6 Coarse Positioning Execution Decision and Distance Calculation

[0095] Based on the corrected angle information, a steering decision mechanism is established for the coarse positioning stage. The steering decision is based on the deviation analysis between the current pointing angle and the target angle; a steering operation is performed when the deviation exceeds a preset threshold. The threshold setting considers the relationship between the current field of view and radar accuracy, and adopts adaptive threshold calculation.

[0096]

[0097] Where α is the basic threshold coefficient, typically ranging from 0.15 to 0.25. For the target distance, For reference distance.

[0098] Simultaneously, the three-dimensional distance between the target and the observation station is calculated, providing important parameters for subsequent zoom strategies:

[0099]

[0100] Where, d 3D is the three-dimensional Euclidean distance, which is the straight-line distance from the origin (0, 0, 0) to the target point (E, N, U) in a three-dimensional Cartesian coordinate system.

[0101] After this step is completed, the output includes the target pointing angle of the photoelectric device after multiple coordinate transformations and error compensation, the accurate three-dimensional distance of the target, the turning execution status, and the accuracy evaluation results of the angle calculation, providing high-precision initial positioning information for photoelectric confirmation and status judgment in step two.

[0102] Step Two: Photoelectric Confirmation and Status Assessment

[0103] The core objective of this step is to establish an intelligent state machine control framework, addressing the fundamental problems of single control strategies and inappropriate responses in traditional fusion systems, and achieving a gradual transition from coarse radar positioning to precise photoelectric tracking. The inputs are the photoelectric ball pointing angle, target distance, and photoelectric image data from step one.

[0104] 2.1 Design Principles of Three-Stage State Machines

[0105] The fundamental flaw of traditional radar-electro-optical fusion systems lies in their adoption of a unified control strategy, which ignores the phased characteristics of the target detection process. The radar guidance phase requires a rapid response to acquire the target, while the precise tracking phase demands stable control to avoid loss. A single threshold cannot satisfy these conflicting needs. Based on the inherent laws of target detection, this invention innovatively divides the entire process into three states with different control requirements:

[0106] The initial detection state design is for scenarios where the radar detects a target for the first time. At this stage, the angle information provided by the radar has a positioning error on the order of meters. Under the narrow field of view conditions of the optoelectronic device, an excessively small turning threshold will cause the optoelectronic sphere to fail to respond to radar guidance, resulting in target search failure. Therefore, a relaxed turning threshold strategy is adopted, set to a relatively large proportional coefficient of the current field of view. ,Right now The value is typically between 0.15 and 0.25. This design allows the electro-optical sphere to respond quickly to changes in radar angle, improving the probability of target acquisition.

[0107] The target detection state design is based on the need for information accumulation and verification. After the optoelectronic device initially detects a target, it needs to verify the target's authenticity through multiple frames of information to avoid erroneous tracking caused by false detections in a single frame or transient interference. During this stage, turning and zooming operations are paused, focusing on information accumulation and target confirmation. The mathematical basis of this design is Bayesian probability updating, which increases the posterior probability of the target's presence through multiple independent observations.

[0108]

[0109] Where P(target | observations) is the posterior probability, P(observations | target) is the likelihood, P(target) is the prior probability, and P(observations) is the evidence, i.e., the total probability of observing the evidence of observations.

[0110] The tracking state design is for scenarios where the target has been identified and precise tracking is required. At this point, the target's authenticity has been verified, and the system's primary task is to maintain a stable lock on the target. An aggressive turning threshold strategy is employed, set as a small proportional coefficient of the current field of view. ,Right now:

[0111]

[0112] in, Typically, the FOV is taken as 0.05-0.15. current This represents the current instantaneous field of view.

[0113] This design ensures that the photoelectric ball responds with high sensitivity to changes in the target's position, preventing tracking loss due to response delay.

[0114] 2.2 Design of Multi-Counter Fusion Decision Mechanism

[0115] A single decision condition is susceptible to transient disturbances, leading to false state transitions and system instability. This invention innovatively designs a multi-counter cross-validation mechanism, achieving reliable state transition decisions through the collaborative work of multiple independent counters.

[0116] Photoelectric counter It records the number of consecutive frames in which no photoelectric target is detected. Its design principle is based on the assumption of continuity in target detection: a real target should not completely disappear within a short period of time. When When the threshold is exceeded, it indicates that the current field of view setting may be incorrect or the target has left the field of view, requiring a zoom operation or state rollback to be triggered. The threshold setting takes into account the frame rate of the optoelectronic device and the target motion characteristics, and is usually set to 3-8 frames.

[0117] Continuous success counter It records the number of frames in which photoelectric targets are successfully detected consecutively. Its design is based on the idea of ​​statistical testing: the probability of multiple consecutive successful detections... Where p is the probability of a single successful detection, and n is the number of consecutive detections. When the number of consecutive successful detections reaches a preset threshold... When the confidence level of the target's existence is high enough, a transition from target detection to tracking can be triggered. The threshold is typically set between 5 and 15 frames to balance detection reliability and response speed.

[0118] Stable tracking counter The system records the number of stable frames during tracking to determine whether the system has entered a stable tracking phase. The criterion for stable tracking is the stable appearance of the target near the center of the field of view, with the variance of the target center coordinates being less than a preset threshold. When the number of stable frames reaches the threshold, a telephoto mode switch is triggered to further improve tracking accuracy.

[0119] Tracking Loss Counter Records the number of consecutive frames in tracking mode where the target is lost. Tracking loss may be caused by factors such as rapid target maneuvering, occlusion, or system response delay. When the number of lost frames exceeds a threshold... When this happens, it indicates that the current tracking strategy has failed and it is necessary to exit the tracking state and search for the target again.

[0120] 2.3 Dynamic Threshold Adaptive Adjustment Algorithm

[0121] Traditional fixed thresholds cannot adapt to the changing needs of different scenarios. This invention designs a dynamic threshold adaptive adjustment mechanism that adjusts the thresholds of each counter in real time based on the current field of view, target distance, and environmental conditions. The dynamic adjustment formula for the turning threshold is:

[0122]

[0123] Where α(state) is the state-related fundamental coefficient, and FOV(t) is the current field of view angle. This is the distance correction factor. This is an environmental correction factor. The distance correction factor is designed based on the inverse relationship between angular resolution and distance: ,in For reference distance, This represents the current target distance.

[0124] 2.4 Photoelectric target matching and trajectory prediction

[0125] The photoelectric target matching algorithm is responsible for associating the YOLO detection results with the expected target. In the initial detection state, due to the lack of historical trajectory information, a field-of-view center distance matching strategy is adopted. The matching score is calculated as follows:

[0126]

[0127] Where w1, w2, and w3 are weighting coefficients. To detect the distance from the center of the bounding box to the center of the image, d max where c is the maximum distance and c is the detection confidence level. This is the size rationality factor.

[0128] 2.5 State Transition Logic and Execution

[0129] Anti-drone electro-optical ball systems face severe false alarm interference problems in the initial stages of operation. Static or dynamic objects such as buildings, trees, and birds within the field of view of the electro-optical equipment may be misidentified as drone targets by the target detection algorithm, generating a large number of false detection results. Traditional systems lack effective false alarm suppression mechanisms, easily getting bogged down in ineffective tracking of false targets, which seriously affects the actual control performance of the system.

[0130] This invention innovatively designs a state transition mechanism based on radar prior verification, ensuring that only targets guided and successfully matched by radar can trigger a state transition in the system. The core idea of ​​this design is to utilize the radar's airspace coverage capability and relatively low false alarm rate to provide reliable prior information for photoelectric detection, fundamentally solving the false alarm problem of photoelectric systems.

[0131] The initial detection of a target and the transition of the detection state must meet the radar-electro-optical dual verification conditions. The transition decision is based on spatial consistency verification and temporal synchronization verification.

[0132] Spatial position consistency verification is determined by calculating the deviation between the photoelectric detected target and the radar predicted position. Considering radar positioning errors and pixel quantization errors in photoelectric detection, a fault-tolerant matching mechanism is established:

[0133]

[0134] Where, d sp For spatial bias, (x opt y opt (x) represents the photoelectric detection coordinates. rad y rad () represents the radar projection coordinates; if they exist... The match is successful.

[0135] in This represents the standard deviation of radar positioning error. To quantize the error in photoelectric detection, To provide a safety margin, a value typically taken as 2 to 3 times the confidence interval is used. The pixel size.

[0136] Time window matching verification ensures the time synchronization of radar and electro-optical data, avoiding erroneous correlations caused by data delays.

[0137] If it exists Then the time matching condition is met.

[0138] Where Δt is the absolute time difference between the timestamps of the photoelectric sensor data and the radar sensor data, t opt For photoelectric timestamps, t rad For radar timestamps, The maximum time deviation that the system can tolerate is usually set to 2 to 3 times the system sampling period.

[0139] The target detection and tracking state transition is based on a cumulative confidence assessment using continuous validation. To avoid false positives caused by a single, accidental match, a confidence accumulation mechanism based on Bayesian updates is established:

[0140]

[0141] Where P(target | obs) is the posterior probability, P(obs | target) is the likelihood probability, P(target) is the prior probability, and P(obs) is the marginal probability of the evidence.

[0142] continuous The cumulative confidence level after each successful verification is calculated as follows:

[0143]

[0144] Among them, P acc To accumulate confidence, P single N represents the success rate of a single validation attempt. ver This represents the number of consecutive successful verifications.

[0145] when The state transition is triggered at time, where It is usually set to 0.95 to correspond to a 95% confidence level.

[0146] The tracking state maintenance and exit mechanism is based on target persistence assessment and system fault tolerance analysis. Considering the potential for temporary occlusion, rapid maneuvers, or occasional failures of the detection algorithm in real-world targets, a hierarchical fault tolerance mechanism is established:

[0147] Level 1 fault tolerance: Maintain tracking status

[0148] Level 2 fault tolerance: Degrade search mode

[0149] Exit tracking: Return to initial detection status

[0150] in For consecutive frame loss, and These are the first and second level fault tolerance thresholds, determined based on the target motion characteristics and system response capabilities.

[0151] The execution of state transitions employs an atomic operation mechanism, ensuring the consistency and uninterrupted nature of the state switching process. This design guarantees the stability of the system during state transitions and avoids state inconsistencies caused by asynchronous operations. Through a radar prior verification mechanism, false alarm interference from the electro-optical system is effectively suppressed, significantly improving the reliability and practicality of the anti-drone system.

[0152] Step 3: Adaptive zoom control

[0153] The core objective of this step is to resolve the fundamental contradiction faced by electro-optical spheres in anti-drone applications: a large field of view is beneficial for target search but lacks detection accuracy, while a small field of view offers high detection accuracy but makes it easy to lose targets. Traditional systems use a fixed focal length setting, which cannot adapt to changes in target distance and differences in the detection stage, resulting in poor detection performance for distant targets and easy loss of tracking for close targets.

[0154] 3.1 Design Concept of Dual-Mode Zoom Strategy

[0155] This invention proposes a dual-mode adaptive zoom strategy, which dynamically selects the zoom mode based on the current system state and target characteristics. The short-focus mode is designed for the search phase, aiming to compensate for radar positioning errors; the long-focus mode is designed for the recognition phase, aiming to ensure that the target reaches a recognizable pixel size.

[0156] The core design principle of the short-focus mode is "fault-tolerant coverage." The position error caused by the angular error of the radar system at the target distance d is:

[0157]

[0158] Where, ε pos The linear position error is given by σ, where d is the slant distance to the target, and σ is the linear position error.az This represents the azimuth error. Considering the target's maneuverability and altitude uncertainty, the total error range that the optical television field needs to cover is:

[0159]

[0160] in For the safety factor, ε pos To predict position error, For maneuver error, This is for altitude error. The field of view in short focal length mode is calculated as follows:

[0161]

[0162] Among them, FOV short For short focal length field of view, R cover This represents the coverage radius.

[0163] The core design philosophy of telephoto mode is "pixel optimization." Target recognition requires a minimum number of pixels N. min Based on the principles of optical imaging, the required focal length is:

[0164]

[0165] in For the target desired number of pixels, For pixel size, The target size is [size]. The corresponding field of view is:

[0166]

[0167] Among them, FOV long For telephoto field of view, w sensor f is the width of the sensor target surface. req The required equivalent focal length.

[0168] 3.2 Timing control design for zoom execution

[0169] Zooming cannot be performed simultaneously with head rotation due to the mechanical characteristics of the photoelectric sphere and limitations of the control system. Zooming during head rotation can cause unpredictable changes in the target's position within the field of view, potentially causing the target to move out of the field of view and resulting in tracking failure.

[0170] Design an operation timing control mechanism to ensure that zooming is performed only after the head has stabilized:

[0171] If it exists Zooming is then allowed. Among them... This is the servo system stabilization time. Simultaneously, the minimum operation interval T is set. interval Prevent command conflicts:

[0172] When satisfied Then the operation is allowed.

[0173] 3.3 Progressive zoom mechanism in tracking mode

[0174] In tracking mode, the system employs a "stabilize first, then refine" zoom strategy. Upon initially entering tracking mode, it waits for N... stable The system ensures stable tracking within a single frame, then triggers a telephoto switch. A periodic evaluation mechanism will be subsequently established, performing evaluations every N frames. update Frame-based evaluation of zoom parameters.

[0175] This design adapts to dynamic changes in target distance, maintaining optimal zoom regardless of whether the target is moving away or closer.

[0176] 3.4 Zoom Recovery Strategy After Target Loss

[0177] Considering the difference between the YOLO detection frequency and the radar update frequency (typically 25Hz vs 10Hz), target loss determination is based on timing analysis. If N frames (usually 5 frames) of radar data are received consecutively without photoelectric detection, tracking loss is determined.

[0178] The recovery strategy employs a three-stage process:

[0179] Waiting phase: Pause zoom Nwait frames to avoid accidental focus adjustments due to temporary obstruction.

[0180] Adjustment phase: Assess whether the current FOV is suitable for a new search.

[0181] Recovery phase: Target recapture is performed in conjunction with the state machine rollback.

[0182] The strategy for zoom adjustment is as follows:

[0183]

[0184] Among them, FOV recovery To restore the field of view, FOV last The field of view (FOV) at the previous moment. search For the search field of view.

[0185] 3.5 Calculation of zoom parameters for multi-constraint optimization

[0186] The determination of actual zoom parameters needs to satisfy hardware constraints, detection constraints, and stability constraints. An optimization objective function is established as follows:

[0187]

[0188] Where w1, w2, and w3 are weighting coefficients, and N opt For the ideal number of pixels, N actualP represents the actual number of pixels. stability As a stability penalty term, f new As a candidate new focal length, f current Given the current focal length, solve for the optimal focal length parameters through constraint optimization.

[0189] This step outputs the determined zoom mode, optimal field of view, zoom execution timing, and expected target pixel count, providing an optimized optoelectronic configuration for precise tracking control in step four.

[0190] Step 4: Precise Tracking Control

[0191] The core objective of this step is to achieve high-precision real-time tracking control of the photoelectric ball after target confirmation, thereby solving the problems of slow response and easy loss of highly maneuverable targets in traditional tracking systems.

[0192] 4.1 Design of Aggressive Turning Strategy

[0193] Traditional photoelectric tracking systems use a fixed head-turn threshold, which cannot meet the accuracy requirements of different tracking stages. This invention uses an aggressive head-turn threshold during tracking, significantly improving tracking accuracy.

[0194] Based on target pixel coordinates (x) target , y target ) and image center (x center , y center ), calculate the angle deviation:

[0195]

[0196] Where, Δ total For the total error or combined error, Δ az For azimuth error, Δ el This represents the pitch angle error.

[0197] The conditions for a radical reversal judgment are:

[0198]

[0199] in This is the aggressive coefficient, typically ranging from 0.03 to 0.08, which is significantly smaller than the threshold for the search phase.

[0200] 4.2 Fine-tuning of telephoto lenses after stable tracking

[0201] System accumulation After frame stabilization tracking, based on the target pixel size Perform telephoto fine adjustment To ensure system stability, smooth control is introduced:

[0202]

[0203] Among them, S pixel As the reference for the target size, w bbox Detection frame width, h bbox f is the height of the detection frame. fine For an ideal fine focal length, f current N is the current focal length. opt For the optimal number of pixels, S pixel f is the current size of the target. adj To adjust the back focal length, k smooth This is a smoothing coefficient, typically ranging from 0.1 to 0.3.

[0204] 4.3 Target Loss Detection and Reacquisition

[0205] Target loss determination is based on continuity analysis; when the number of consecutively lost frames C... loss >T loss Recapture is triggered at certain times. The recapture strategy employs a tiered processing approach:

[0206] Short-term loss: Maintain the current direction and wait for the target to reappear.

[0207] Long-term loss: Expand the search or backtrack the state machine based on motion prediction

[0208] Motion prediction uses linear extrapolation of historical trajectories:

[0209]

[0210] Where, p pred To predict the position vector, p last v is the last known position vector. avg The average velocity vector; t lost The time lost for the target.

[0211] 4.4 Adaptive Control Parameter Optimization

[0212] Adjust control parameters in real time based on target motion characteristics:

[0213]

[0214] Where, x adapt For adaptive parameters, x base Based on the basic parameter value, k motion v is the motion coupling coefficient. target The target speed.

[0215] Establish a tracking quality assessment:

[0216]

[0217] Among them, Q trackTo track the overall quality score, w1, w2, and w3 are weighting coefficients, and C conf To identify the confidence level, S stable For stability scoring, Δtotal represents the total deviation, which is the total angle of deviation between the target's imaging position and the center of the image, as mentioned earlier. FOV current This is the current field of view.

[0218] This step outputs precise electro-optical ball control commands, tracking quality assessments, and target motion prediction information, enabling high-precision and stable tracking of UAV targets.

[0219] The above steps enabled the transformation of training data based on detection results. By introducing mechanisms such as area threshold filtering, frame interval control, and normalized coordinate calculation, the accuracy and consistency of the generated data were ensured. Furthermore, the introduction of manual judgment made the generation of ship category labels more reliable, providing a high-quality data foundation for subsequent training of YOLO-based deep learning models.

[0220] Working Principle: The system achieves intelligent tracking of UAV targets through a three-stage state machine framework comprising initial detection, target detection, and tracking states. Upon system startup, it first enters the initial detection state. In this state, the radar provides the initial target orientation, and the electro-optical ball uses a relaxed control strategy to quickly point towards the approximate target area, ensuring a high initial acquisition success rate. When the target enters the field of view, the system transitions to the target detection state. At this point, large-scale movements are paused, and a multi-counter mechanism verifies the electro-optical information across multiple frames to ensure target authenticity and provide a reliable basis for state transitions. After target confirmation, the system enters the high-precision tracking state. An aggressive control strategy is employed to respond quickly to minor target deviations. Simultaneously, an adaptive zoom algorithm dynamically adjusts the focal length based on target distance and recognition requirements, balancing search range and recognition accuracy. Throughout the process, the system utilizes underlying technologies such as coordinate transformation, error compensation, motion prediction, and quality assessment to ensure the accuracy and stability of the tracking process, ultimately achieving intelligent control of the entire process from rapid acquisition to stable tracking of UAV targets.

[0221] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, 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 process, method, article, or apparatus.

[0222] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A multi-stage state machine based opto-electronic ball drone tracking control method, characterized in that, The method comprises the following steps: Step one, radar-guided coarse positioning: receiving radar target information, calculating the target pointing angle of the photoelectric sphere through coordinate system conversion and error compensation; Step two, photoelectric confirmation and state judgment: establishing a three-stage state machine including initial detection, target detection and tracking state, realizing dynamic switching between states according to the matching results of radar and photoelectric data, and combining a multi-counter fusion decision mechanism; The switching logic of the three-stage state machine is as follows: in the initial detection state, a loose steering threshold is used to quickly respond to radar guidance; when the target in the initial detection state is successfully verified by the spatial and temporal consistency of radar and photoelectric data, the target detection state is switched to, and the steering and zoom operations are suspended to accumulate verification information; When the continuous success counter in the target detection state reaches a first preset threshold, the tracking state is switched to; In the tracking state, a radical steering threshold is used to keep the target locked, and if the number of consecutive lost frames exceeds a second preset threshold, the initial detection state is returned to; Step three, adaptive zoom control: based on the target distance, radar positioning error and preset pixel requirements, a dual-mode zoom strategy is used to dynamically adjust the field of view angle of the photoelectric device; Step four, precise tracking control: in the tracking state, the aggressive steering threshold strategy based on angle deviation is adopted, and the aggressive steering threshold θ threshold in the strategy is a dynamic value, combined with target motion prediction and tracking quality evaluation Q track , to achieve high-precision stable tracking.

2. The photovoltaic ball unmanned aerial vehicle tracking control method based on a multi-stage state machine according to claim 1, characterized in that, The multi-counter fusion decision mechanism of step two comprises: An optical counter is used to record the number of frames in which the photoelectric target is not detected continuously, and when the value exceeds a third preset threshold, an alarm or state rollback is triggered; A continuous success counter is used to record the number of frames in which the photoelectric and radar targets are successfully matched continuously, which is used as the main decision condition for switching from the target detection state to the tracking state; A stable tracking counter is used to record the number of stable frames in which the target is in the center area of the field of view in the tracking state, which is used as a condition for triggering long focus mode switching.

3. The photoball drone tracking control method based on multi-stage state machine according to claim 1, wherein, The dual-mode zoom strategy of step three comprises: A short focus mode, whose field of view angle is calculated based on the coverage radius synthesized by radar positioning error, target maneuvering error and height error, and target distance, is used to ensure that the target uncertainty area is covered in the coarse positioning stage; A long focus mode, whose focal length is calculated based on the focal length required to make the target reach a preset optimal number of pixels in the image, is used to provide high-precision imaging in the tracking stage.

4. The photoball drone tracking control method based on multi-stage state machine according to claim 3, characterized in that, The field of view angle of the short focus mode is calculated by the following formula: ； wherein R cover is the coverage radius, obtained by taking the square root of the sum of the squares of the position error, the maneuver error, and the altitude error, multiplied by a safety factor, and d is the target slant range.

5. The photoball drone tracking control method based on multi-stage state machine according to claim 1, wherein, The coordinate system conversion of step one comprises: Converting the target geographic coordinate system (λ, φ, h) provided by the radar into the geocentric rectangular coordinate system (X, Y, Z); Converting the geocentric rectangular coordinate system into the station-centered east-north-up coordinate system (E, N, U) with the observation station as the origin; Based on the station-centered coordinate system, the azimuth and elevation angles required by the photoelectric sphere are calculated, and atmospheric refraction correction is added.

6. The multi-stage state machine based photo-ball UN tracking control method according to claim 1, wherein, For error compensation of step one, a linear model including installation deviation, temperature drift correction and gravity deformation correction is used to compensate the calculated azimuth and elevation angles to obtain the final control command angle.

7. The multi-stage state machine based photo-ball UN tracking control method according to claim 1, wherein, the aggressive turn threshold θ of step four threshold is calculated by the following equation: ； wherein, Typically 0.05-0.15, FOV current is the current instantaneous field of view angle.

8. The multi-stage state machine based photo-ball UN tracking control method according to claim 1, wherein, The target motion prediction of step four uses a linear extrapolation method based on the last known position and average velocity vector of the target before loss to predict the possible position during the loss period for target recapture.

9. The multi-stage state machine based photo-ball UN tracking control method according to claim 1, wherein, The tracking quality assessment Q is determined for step four track The confidence C is identified by a weighted calculation conf The stability score S stable and the position score, which is one minus the ratio of the total deviation of the target in the field of view and the current field of view angle, the assessment result is used to trigger adaptive adjustment of control parameters or switching of tracking strategies.

Images