Anti-robot narrowband internet of things (NB-IoT) follow-up cloud platform automatic tracking method

Abstract

The application discloses a countermeasure robot follow-up type cloud platform automatic tracking method based on narrowband Internet of Things, relates to the technical field of robot control, and comprises the following steps: step one: collecting target observation data, link data, robot posture, cloud platform field of view parameters and cloud platform state, forming a tracking data set; step two: performing time sequence registration and coordinate mapping, generating a pointing error state; step three: constructing an improved Ornstein-Uhlenbeck model and an action completion time length distribution; step four: constructing a terminal capture domain, generating a terminal capture distribution; step five: generating a terminal capture potential function, reconstructing a drift term; step six: generating an effective message arrival distribution and a mobile field of view safety domain; step seven: generating a communication silence window survival function, updating the drift term; and step eight: generating a target line of sight prediction trajectory and a cloud platform control amount, and driving the cloud platform to track the target. The application improves tracking stability, accuracy and anti-time delay capability.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and in particular to an automatic tracking method for countermeasure robots using a gimbal-type servo system based on narrowband Internet of Things (IoT). Background Technology

[0002] With the increasing application of countermeasure robots in low-altitude target inspection, area surveillance, and mobile response scenarios, the technology of transmitting target observation information and automatically tracking the gimbal using narrowband IoT links has attracted widespread attention. Existing countermeasure robots mainly calculate target position deviations based on image detection results and control gimbal rotation through fixed delay compensation and linear prediction. However, the following problems commonly exist in practical applications: The posture changes caused by robot movement will create dynamic deviations between image coordinates and gimbal coordinates. The lack of unified temporal registration of target observation data, robot posture, and gimbal status leads to inaccurate calculation of target line-of-sight deviation and change rate, and the gimbal is prone to tracking lag and oscillation. Narrowband IoT links suffer from bandwidth limitations, message retransmission, and buffer queuing. The communication completion time varies with signal reception power and link status. Traditional fixed delay compensation is difficult to characterize the action completion time distribution formed by the communication completion time and the gimbal mechanical response time. When the target moves rapidly and the communication silent window is superimposed, traditional linear extrapolation methods cannot simultaneously consider the gimbal's reachable range and field of view boundary. The predicted target line-of-sight trajectory is prone to exceed the moving field of view safety domain, causing target loss and reducing the stability and accuracy of continuous tracking by the servo gimbal.

[0003] Therefore, how to provide an automatic tracking method for countermeasure robots based on narrowband IoT is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose an automatic tracking method for a countermeasure robot servo gimbal based on narrowband Internet of Things (IoT). This invention fully utilizes narrowband IoT link analysis, an improved Ornstein-Uhlenbeck model, and gimbal predictive control technology to achieve target line-of-sight prediction and continuous gimbal tracking. It has the advantages of strong anti-communication latency capability, high tracking stability, and high control accuracy.

[0005] The automatic tracking method for countermeasure robots using a gimbal based on narrowband Internet of Things according to an embodiment of the present invention includes the following steps: Step 1: Collect target observation data and narrowband IoT link data to obtain robot posture, gimbal field of view parameters and gimbal status, forming a tracking dataset; Step 2: Perform temporal registration and coordinate mapping on the tracking dataset to generate a pointing error state consisting of target line-of-sight deviation and rate of change; Step 3: Construct an improved Ornstein-Uhlenbeck model with the pointing error state, and generate the action completion time distribution from the tracking dataset; Step 4: Construct the terminal capture domain based on the action completion time distribution and gimbal status, and generate the terminal capture distribution; Step 5: Based on the action completion time distribution and terminal capture distribution, process the model transition density, generate the terminal capture potential function, and reconstruct the drift term from the state gradient of the terminal capture potential function; Step 6: Generate valid messages from the tracking dataset to reach the distribution, and construct a mobile field of view safety domain based on the pointing error status and PTZ field of view parameters; Step 7: Calculate the first exit time of the pointing error state when it leaves the safe domain of the moving field of view, generate the communication silent window survival function by combining the distribution of valid message arrivals, and update the drift term by the state gradient of the communication silent window survival function. Step 8: Solve the model according to the drift term, generate the target line-of-sight prediction trajectory and convert it into gimbal control quantity to drive the gimbal to track the target.

[0006] Optionally, step one specifically includes: Collect the target center pixel coordinates and target bounding box scale and record the target sampling time. Correlate the target center pixel coordinates and target bounding box scale with the target sampling time to form target observation data. Record the sending and receiving times of narrowband IoT messages, collect the number of message retransmissions, signal receiving power, and buffer queuing time to form narrowband IoT link data; Collect robot heading angle, robot pitch angle and robot roll angle to form robot posture, obtain gimbal horizontal field of view angle and gimbal vertical field of view angle to form gimbal field of view parameters, collect gimbal azimuth angle, gimbal pitch angle, gimbal angular velocity and gimbal angular acceleration to form gimbal state. Based on the target sampling time, message sending time, and message receiving time, target observation data and narrowband IoT link data are associated, and robot posture, gimbal field of view parameters, and gimbal status are bound to the target observation data to form a tracking dataset.

[0007] Optionally, step two specifically includes: The tracking dataset is sorted according to the target sampling time, and the target center pixel coordinates are associated with the corresponding robot posture and gimbal status to generate a time-series registration record. Calculate the horizontal and vertical pixel offsets of the target center pixel coordinates relative to the imaging center, and convert the horizontal and vertical pixel offsets into the target line-of-sight azimuth and target line-of-sight pitch angles according to the gimbal field of view parameters to generate the target line-of-sight angle; The robot's posture is converted into a posture rotation matrix. The posture rotation matrix is ​​used to perform coordinate rotation on the target line of sight angle. The coordinate rotation result is compared with the gimbal azimuth and gimbal pitch angles in the gimbal state to calculate the angle difference and generate the target line of sight deviation. Calculate the difference in target line-of-sight deviation and the difference in target sampling time corresponding to consecutive target sampling times. Divide the difference in target line-of-sight deviation by the difference in target sampling time to generate the rate of change. Combine the target line-of-sight deviation and the rate of change according to the target sampling time to generate the pointing error state.

[0008] Optionally, step three specifically includes: Arrange the pointing error states according to the target sampling time, combine the target line-of-sight deviation and the rate of change into a state vector sequence, and calculate the mean regression center and the increment of adjacent state vectors of the state vector sequence. Using the deviation of the state vector from the mean regression center as the independent variable and the increment of the adjacent state vector as the dependent variable, least squares fitting is performed to generate the drift matrix. The fitting error covariance is calculated and decomposed to generate the diffusion matrix. The mean regression center, drift matrix, and diffusion matrix are written into the Ornstein-Uhlenbeck state equation to construct the improved Ornstein-Uhlenbeck model and generate the model transition density. Extract the target sampling time, message reception time, message retransmission count, signal reception power, and PTZ status from the tracking dataset. Subtract the target sampling time from the message reception time to generate the communication completion time. Calculate the PTZ mechanical response time when the PTZ angular displacement reaches the target line-of-sight deviation based on the target line-of-sight deviation and PTZ status. The action completion time sample is generated by adding the communication completion time and the gimbal mechanical response time. The action completion time sample is divided according to the message retransmission count and the signal received power range. The frequency of occurrence of the action completion time sample in each range is counted and the probability is normalized to generate the action completion time distribution.

[0009] Optionally, step four specifically includes: Extract the action completion time interval and interval probability from the action completion time distribution, and use the gimbal azimuth angle, gimbal pitch angle, gimbal angular velocity and gimbal angular acceleration to calculate the azimuth reachable angle, pitch reachable angle, azimuth reachable velocity and pitch reachable velocity corresponding to the action completion time interval. The target line-of-sight deviation is compared with the azimuth reachable angle and the pitch reachable angle, and the rate of change is compared with the azimuth reachable speed and the pitch reachable speed. Pointing error states that meet the angle reachable condition and the speed reachable condition are selected to generate the capture subdomain. The capture subdomains are merged according to the action completion time interval, and the interval probability is configured to the corresponding capture subdomain to generate a terminal capture domain consisting of the target line-of-sight deviation range, the change speed range, and the interval probability. Calculate the target line-of-sight deviation amplitude and change rate amplitude corresponding to each pointing error state within the terminal capture domain. Generate state weights based on the target line-of-sight deviation amplitude and change rate amplitude. Multiply the state weights by the interval probability and perform probability normalization to generate the terminal capture distribution.

[0010] Optionally, step five specifically includes: Extract the action completion time interval and interval probability from the action completion time distribution, and extract the state probability corresponding to the pointing error state in the terminal capture domain from the terminal capture distribution; Using the current pointing error state as the starting state and the pointing error state in the terminal capture domain as the ending state, the action completion time interval is substituted into the model transition density to calculate the state transition probability from the starting state to the ending state. Multiply the state transition probability, state probability, and interval probability, sum them according to the terminal capture domain and the action completion time interval to generate the terminal capture potential function. Calculate the partial derivative of the terminal capture potential function with respect to each dimension of the pointing error state to generate the state gradient of the terminal capture potential function. Multiply the diffusion matrix by its transpose, then multiply the product matrix by the gradient of the terminal capture potential function state to generate the terminal capture drift increment. This terminal capture drift increment is then added to the drift matrix and applied to the base drift amount formed by the difference between the pointing error state and the mean regression center, thus completing the drift term reconstruction.

[0011] Optionally, the model transition density takes the initial pointing error state, the final pointing error state, and the state transition duration as inputs. It calculates the state transition mean based on the mean regression center, the drift matrix, and the initial pointing error state. It calculates the state transition covariance based on the drift matrix, the diffusion matrix, and the state transition duration. It calculates the probability density based on the state deviation between the final pointing error state and the state transition mean, as well as the state transition covariance. This density is used to characterize the probability density of the initial pointing error state transitioning to the final pointing error state within the state transition duration.

[0012] Optionally, step six specifically includes: Extract the target sampling time and message reception time from the tracking dataset, perform time-series matching between the message reception time and the target sampling time, retain the messages that are associated with the target observation data, and generate valid message records; Arrange valid message records according to message reception time, calculate the difference between reception times of adjacent messages, generate valid message arrival time samples, divide valid message arrival time samples according to message retransmission count and signal reception power range, count the occurrence frequency of valid message arrival time samples and perform probability normalization to generate valid message arrival distribution. Extract the target line-of-sight deviation and rate of change from the pointing error state, and determine half of the horizontal field of view of the gimbal as the azimuth field of view boundary and half of the vertical field of view of the gimbal as the pitch field of view boundary. The arrival time intervals in the valid message arrival distribution are multiplied by the rate of change to generate a predicted offset. The predicted offset is added to the target line-of-sight deviation. The pointing error state of the summation result within the azimuth and pitch field-of-sight boundaries is retained. The pointing error states are then merged according to the arrival time intervals and interval probabilities to generate a moving field-of-sight safety domain.

[0013] Optionally, step seven specifically includes: Using the current pointing error state as the initial state, the pointing error state is advanced by the reconstructed drift term and diffusion matrix of the improved Ornstein-Uhlenbeck model. The pointing error state at each advancement moment is compared with the boundary of the moving field of view safety domain. The advancement moment when the pointing error state first crosses the boundary of the moving field of view safety domain is determined as the first exit time. The arrival time interval and interval probability are extracted from the valid message arrival distribution. The survival condition is that the first exit time is not less than the upper limit of the arrival time interval. The survival probability corresponding to the survival condition is calculated based on the model transition density. The survival probability is multiplied by the interval probability and summed according to the arrival time interval to generate the communication silent window survival function. Perform a natural logarithmic transformation on the communication silent window survival function, calculate the partial derivatives of the natural logarithmic transformation result with respect to each dimension of the pointing error state, and generate the state gradient of the communication silent window survival function. Multiply the diffusion matrix by its transpose, then multiply the product matrix by the state gradient of the communication silent window survival function to generate the field-of-view survival drift increment. Finally, add the field-of-view survival drift increment to the reconstructed drift term to complete the drift term update.

[0014] Optionally, step eight specifically includes: Using the current pointing error state as the initial state, the updated drift term and diffusion matrix are substituted into the improved Ornstein-Uhlenbeck model. The pointing error state is advanced within each action completion time interval according to the target sampling time interval, generating the predicted pointing error state sequence corresponding to the action completion time interval. The target line-of-sight deviation is extracted from the predicted pointing error state sequence. The target line-of-sight deviation is then superimposed with the gimbal azimuth and gimbal pitch angles to generate a target line-of-sight angle sequence corresponding to the action completion time interval. The target line-of-sight angle sequence is weighted and summarized using interval probability to generate the target line-of-sight prediction trajectory. Calculate the angle difference between adjacent target line-of-sight angles and the difference in target sampling time in the target line-of-sight prediction trajectory. Divide the angle difference by the difference in target sampling time to generate the gimbal angular velocity. Combine the target line-of-sight angle and gimbal angular velocity in the target line-of-sight prediction trajectory to generate the gimbal control quantity. Input the gimbal control parameters into the gimbal to control the azimuth and pitch angles of the gimbal to change according to the predicted trajectory of the target's line of sight, and adjust the rotation process according to the angular velocity of the gimbal to drive the gimbal to track the target.

[0015] The beneficial effects of this invention are: This invention establishes a pointing error state composed of target line-of-sight deviation and change rate by performing temporal registration and coordinate mapping on target observation data, narrowband IoT link data, robot posture, gimbal field-of-sight parameters, and gimbal status. An improved Ornstein-Uhlenbeck model is then used to describe the stochastic evolution of the target line-of-sight state. By constructing the action completion time distribution through communication completion time and gimbal mechanical response time, and combining this with gimbal reachability angle, gimbal reachability speed, and terminal acquisition distribution, a terminal acquisition potential function is generated. This allows for the unified incorporation of link transmission delay, gimbal mechanical response capability, and target motion state into the prediction process, reducing tracking lag and reciprocating oscillations caused by fixed delay compensation and linear extrapolation, and improving the matching degree between the target line-of-sight prediction results and the actual execution capability of the gimbal.

[0016] This invention further constructs a mobile field-of-view safety domain based on the distribution of valid message arrivals. The drift term is updated using the first exit time and the communication silent window survival function, ensuring that the improved Ornstein-Uhlenbeck model can still constrain the pointing error state within the gimbal's field of view even under conditions of increased narrowband IoT message intervals, message retransmissions, and changes in signal received power. The updated drift term is used to generate the target line-of-view prediction trajectory and convert it into gimbal control variables. This maintains continuous gimbal rotation and stable target tracking in environments with robot posture changes and communication link fluctuations, reducing tracking interruptions caused by the target going out of the field of view, and improving the stability, accuracy, and resilience to communication latency of the servo gimbal's automatic tracking. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1The flowchart shows the automatic tracking method of the countermeasure robot servo gimbal based on narrowband Internet of Things proposed in this invention. Figure 2 This is a flowchart illustrating the construction of an improved Ornstein-Uhlenbeck model for the automatic tracking method of a countermeasure robot based on narrowband Internet of Things proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-2 The automatic tracking method for countermeasure robots using a servo gimbal based on narrowband IoT includes the following steps: Step 1: Collect target observation data and narrowband IoT link data to obtain robot posture, gimbal field of view parameters and gimbal status, forming a tracking dataset; Step 2: Perform temporal registration and coordinate mapping on the tracking dataset to generate a pointing error state consisting of target line-of-sight deviation and rate of change; Step 3: Construct an improved Ornstein-Uhlenbeck model with the pointing error state, and generate the action completion time distribution from the tracking dataset; Step 4: Construct the terminal capture domain based on the action completion time distribution and gimbal status, and generate the terminal capture distribution; Step 5: Based on the action completion time distribution and terminal capture distribution, process the model transition density, generate the terminal capture potential function, and reconstruct the drift term from the state gradient of the terminal capture potential function; Step 6: Generate valid messages from the tracking dataset to reach the distribution, and construct a mobile field of view safety domain based on the pointing error status and PTZ field of view parameters; Step 7: Calculate the first exit time of the pointing error state when it leaves the safe domain of the moving field of view, generate the communication silent window survival function by combining the distribution of valid message arrivals, and update the drift term by the state gradient of the communication silent window survival function. Step 8: Solve the model according to the drift term, generate the target line-of-sight prediction trajectory and convert it into gimbal control quantity to drive the gimbal to track the target.

[0020] In this embodiment, step one specifically includes: Acquire target image frames, determine the center position and boundary size of the target detection box from the target image frames, convert the center position into target center pixel coordinates, convert the boundary size into target box scale, and use the generation time of the target image frame as the target sampling time. Arrange the target center pixel coordinates and target box scale according to the target sampling time to form target observation data. During narrowband IoT message transmission, the time when the message enters the sending queue and the time when the message arrives at the receiving end are recorded. The time when the message enters the sending queue is determined as the message sending time, and the time when the message arrives at the receiving end is determined as the message receiving time. The number of message retransmissions, signal receiving power, and buffer queuing time corresponding to a single message transmission are also recorded to form narrowband IoT link data. The heading angle, pitch angle and roll angle of the countermeasure robot at the target sampling time are collected and arranged according to the attitude coordinate axis to form the robot attitude. The horizontal field of view and vertical field of view corresponding to the imaging range of the follow-up gimbal are read to form the gimbal field of view parameters. The azimuth angle, pitch angle, angular velocity and angular acceleration of the follow-up gimbal at the target sampling time are collected to form the gimbal state. Using the target sampling time as the correlation benchmark, a correspondence is established between the target observation data and the message sending and receiving times that match the time range. The corresponding narrowband IoT link data is associated with the target observation data, and the robot posture, gimbal field of view parameters and gimbal status are bound according to the target sampling time to form a tracking dataset arranged in chronological order.

[0021] In this embodiment, step two specifically includes: The tracking dataset is sorted in ascending order according to the target sampling time. The target center pixel coordinates, robot posture and gimbal status corresponding to each target sampling time are written to the same time position. Data that is missing the corresponding time position is stopped from being paired. A time registration record is generated that records the correspondence between the target sampling time, the target center pixel coordinates, the robot posture and the gimbal status. The horizontal pixel offset is obtained by subtracting the horizontal coordinate of the target center pixel from the horizontal coordinate of the imaging center, and the vertical pixel offset is obtained by subtracting the vertical coordinate of the target center pixel from the vertical coordinate of the imaging center. The target line-of-sight azimuth angle is obtained by multiplying the ratio of the horizontal pixel offset to the image width by the horizontal field of view of the gimbal, and the target line-of-sight pitch angle is obtained by multiplying the ratio of the vertical pixel offset to the image height by the vertical field of view of the gimbal. The target line-of-sight azimuth angle and the target line-of-sight pitch angle are combined to form the target line-of-sight angle. The robot's heading angle, pitch angle, and roll angle in the robot's posture are converted into an attitude rotation matrix according to the coordinate axis rotation order. The attitude rotation matrix is ​​used to perform coordinate rotation on the target line of sight angle. The azimuth difference between the target line of sight azimuth angle and the gimbal azimuth angle after coordinate rotation is calculated. The pitch difference between the target line of sight pitch angle and the gimbal pitch angle after coordinate rotation is also calculated. The azimuth difference and pitch difference are combined to form the target line of sight deviation. Extract the target line-of-sight deviation from two consecutive sampling times, calculate the changes in azimuth and pitch angles respectively, and calculate the time difference between the two consecutive sampling times. Divide the change in azimuth angle by the time difference to obtain the azimuth change rate, and divide the change in pitch angle by the time difference to obtain the pitch change rate. Combine the azimuth and pitch change rates to form the change rate. Then combine the target line-of-sight deviation and change rate corresponding to the same sampling time to form the pointing error state.

[0022] In this embodiment, step three specifically includes: The pointing error states are arranged in ascending order according to the target sampling time. The target line deviation and change rate corresponding to each target sampling time are written into the same state vector to form a state vector sequence. The arithmetic mean of each component in the state vector sequence is calculated to obtain the mean regression center. The state vector corresponding to the previous target sampling time is subtracted from the state vector corresponding to the next target sampling time to obtain the adjacent state vector increment. The deviation between each state vector in the state vector sequence and the mean regression center is calculated. The deviation is used as the independent variable and the increment of adjacent state vectors is used as the dependent variable. The drift matrix is ​​determined by minimizing the squared difference between the dependent variable and the fitted value. The fitting error between the fitted value and the dependent variable is calculated and the fitting error covariance is formed. The matrix factorization of the fitting error covariance is performed to obtain the diffusion matrix. The mean regression center, drift matrix and diffusion matrix are written into the Ornstein-Uhlenbeck state equation to form an improved Ornstein-Uhlenbeck model. The model transition density is calculated according to the state transition relationship in the state equation. Extract the target sampling time, message reception time, message retransmission count, signal reception power and gimbal status corresponding to the same target observation data from the tracking dataset. Use the time difference between the message reception time and the target sampling time as the communication completion time. Convert the target line-of-sight deviation into the gimbal angular displacement to be completed. Calculate the time required for the gimbal angular displacement to reach the target line-of-sight deviation according to the gimbal angular velocity and gimbal angular acceleration in the gimbal status. Generate the gimbal mechanical response time. The communication completion time and gimbal mechanical response time corresponding to the same target observation data are added together to generate an action completion time sample. The action completion time sample is initially divided according to the number of message retransmissions. Then, the initial division result is further divided according to the signal received power range. The frequency of occurrence of action completion time samples in each division range is counted. The interval probability is obtained by dividing the frequency of occurrence by the total number of action completion time samples. The action completion time distribution is generated according to the correspondence between action completion time and interval probability.

[0023] This invention addresses the problem that the traditional Ornstein-Uhlenbeck model, which uses fixed mean regression parameters and struggles to reflect the coupling relationship between target motion, communication latency, and gimbal mechanical response, combines target line-of-sight deviation and velocity change into a state vector. This allows the model to simultaneously describe azimuth deviation, pitch deviation, and their corresponding trends. A state vector sequence is formed according to the target sampling time. A drift matrix is ​​generated by least-squares fitting between the deviation of the state vector relative to the mean regression center and the increments of adjacent state vectors, allowing the mean regression direction and intensity to be determined based on actual tracking data. A diffusion matrix is ​​generated by matrix decomposition of the fitting error covariance, enabling the correlation between random disturbance amplitude and state dimensions to be jointly characterized by target observation fluctuations and robot posture changes. Simultaneously, the mean regression center, drift matrix, and diffusion matrix are incorporated into the state equation to generate the model transition density, ensuring the current pointing error state... The probability of state evolution towards the target state can be quantified. In terms of time scale processing, the time difference between the message reception time and the target sampling time is determined as the communication completion time. The mechanical execution process formed by the target line-of-sight deviation, gimbal angular velocity, and gimbal angular acceleration is converted into the gimbal mechanical response time. The communication completion time and the gimbal mechanical response time are then combined into an action completion time sample. The action completion time distribution is established based on the number of message retransmissions and signal reception power. This makes the model no longer dependent on a fixed prediction time, and can adapt to narrowband IoT link fluctuations, buffer queuing, message retransmissions, and changes in gimbal response capabilities. It improves the matching degree between the pointing error state prediction and the actual control action completion time, reduces the prediction offset, gimbal tracking lag, and reciprocating oscillation caused by fixed delay compensation, and provides a stable and computable state evolution basis for terminal acquisition potential function calculation, drift term reconstruction, and target line-of-sight prediction trajectory generation.

[0024] In this embodiment, step four specifically includes: Extract the action completion time interval and interval probability from the action completion time distribution. Substitute the lower and upper limits of the action completion time interval into the gimbal angular displacement calculation relationship. Use the gimbal azimuth, gimbal pitch, gimbal angular velocity and gimbal angular acceleration to calculate the azimuth and pitch angles corresponding to the interval endpoints. Use the minimum and maximum values ​​in the interval endpoint calculation results as the boundaries of the azimuth reachable angle and pitch reachable angle, respectively. Use the gimbal angular velocity and gimbal angular acceleration to calculate the azimuth angular velocity and pitch angular velocity corresponding to the interval endpoints to form the azimuth reachable velocity and pitch reachable velocity. The target line-of-sight deviation is decomposed into azimuth deviation and pitch deviation. It is determined whether the azimuth deviation is within the reachable azimuth angle and whether the pitch deviation is within the reachable pitch angle to form the angle reachability condition. The change rate is decomposed into azimuth change rate and pitch change rate. It is determined whether the azimuth change rate is within the reachable azimuth speed and whether the pitch change rate is within the reachable pitch speed. The pointing error state that simultaneously satisfies the angle reachability condition and the speed reachability condition is retained to generate the capture subdomain. The capture subdomains are merged according to the action completion time interval. The interval probabilities corresponding to the action completion time intervals are configured to the capture subdomains generated in the same action completion time interval. The target line-of-sight deviation range is formed by the azimuth deviation boundary and the pitch deviation boundary in the capture subdomain. The change speed range is formed by the azimuth change speed boundary and the pitch change speed boundary. The target line-of-sight deviation range, change speed range and interval probabilities are combined to generate the terminal capture domain. The sum of the squares of the azimuth deviation and pitch deviation in each pointing error state within the terminal acquisition domain is calculated and the square root is taken to obtain the target line-of-sight deviation amplitude. The sum of the squares of the azimuth change rate and pitch change rate is calculated and the square root is taken to obtain the change rate amplitude. The state weight is reduced in the direction of increasing target line-of-sight deviation amplitude and change rate amplitude. The state weight is multiplied by the interval probability of the corresponding acquisition subdomain to obtain the state probability value. The state probability value is then divided by the sum of the state probability values ​​within the terminal acquisition domain to generate the terminal acquisition distribution.

[0025] In this embodiment, step five specifically includes: Extract the lower limit, upper limit and interval probability of each action completion time interval from the action completion time distribution; extract the state probability corresponding to each pointing error state in the terminal capture domain from the terminal capture distribution; and arrange the interval probability and state probability according to the correspondence between the action completion time interval and the terminal capture domain. The pointing error state corresponding to the current target sampling time is determined as the starting state, and each pointing error state in the terminal capture domain is determined as the ending state. The lower limit and upper limit of the action completion time interval are used as the state transition time input model transition density, respectively. The probability of the starting state transitioning to each ending state within the action completion time interval is calculated, and the probability corresponding to the interval endpoint is integrated to generate the state transition probability. Multiply the state transition probability corresponding to each termination state by the state probability, and then multiply it by the interval probability corresponding to the action completion time interval to obtain the capture contribution value of the termination state within the action completion time interval. Summate the capture contribution values ​​within the terminal capture domain and sum them again for the summation results corresponding to the action completion time interval to generate the terminal capture potential function corresponding to the current pointing error state. Sequentially change the azimuth deviation, pitch deviation, azimuth change rate, and pitch change rate in the pointing error state and calculate the rate of change of the terminal capture potential function to generate the state gradient of the terminal capture potential function. Multiply the diffusion matrix with its transpose to form the diffusion covariance matrix. Multiply the diffusion covariance matrix with the state gradient of the terminal capture potential function to form the terminal capture drift increment. Calculate the difference between the error state and the mean regression center and perform a matrix transformation on the difference using the drift matrix to generate the base drift. Add the terminal capture drift increment to the base drift to form the reconstructed drift term.

[0026] In this embodiment, the model transition density takes the initial pointing error state, the final pointing error state, and the state transition duration as inputs. The mean of the state transition is calculated based on the mean regression center, the drift matrix, and the initial pointing error state. The state transition covariance is calculated based on the drift matrix, the diffusion matrix, and the state transition duration. The probability density is calculated based on the state deviation between the final pointing error state and the mean of the state transition, as well as the state transition covariance. This density is used to characterize the probability density of the initial pointing error state transitioning to the final pointing error state within the state transition duration.

[0027] In this embodiment, step six specifically includes: Extract the target sampling time and message reception time from the tracking dataset, arrange the target sampling time and message reception time in ascending order of time, match the message reception time with the target sampling time that is adjacent in time position and earlier than the message reception time, retain the messages that establish a time sequence association with the target observation data, and write the target sampling time and message reception time into the same record to generate a valid message record; Valid message records are sorted in ascending order according to message reception time. The valid message arrival time sample is obtained by subtracting the previous message reception time from the next message reception time. The valid message arrival time sample is initially divided according to the message retransmission count, and then further divided according to the signal reception power range. The frequency of occurrence of valid message arrival time samples in each range is counted. The range probability is obtained by dividing the frequency by the total number of valid message arrival time samples. The valid message arrival distribution is generated according to the correspondence between arrival time ranges and range probabilities. The target line-of-sight deviation and rate of change are extracted from the pointing error state. The target line-of-sight deviation is decomposed into azimuth deviation and pitch deviation, and the rate of change is decomposed into azimuth change rate and pitch change rate. Half of the horizontal field of view of the gimbal is used as the positive and negative boundaries that the azimuth deviation can reach, and half of the vertical field of view of the gimbal is used as the positive and negative boundaries that the pitch deviation can reach, thus forming the azimuth field of view boundary and the pitch field of view boundary. The lower and upper limits of the arrival time interval are multiplied by the azimuth change rate to form the azimuth prediction offset boundary, and the lower and upper limits of the arrival time interval are multiplied by the pitch change rate to form the pitch prediction offset boundary. The azimuth prediction offset boundary is added to the azimuth deviation, and the pitch prediction offset boundary is added to the pitch deviation. The pointing error state that is within the azimuth and pitch field boundaries of the summation result is retained. The pointing error state is merged according to the arrival time interval and interval probability to generate the moving field of view safety domain.

[0028] In this embodiment, step seven specifically includes: Using the pointing error state corresponding to the current target sampling time as the initial state, the advancement time is divided according to the target sampling time interval. At each advancement time, the deterministic state increment is calculated using the reconstructed drift term of the improved Ornstein-Uhlenbeck model, and the random state increment is calculated using the diffusion matrix. The deterministic state increment and the random state increment are superimposed on the pointing error state of the previous advancement time to obtain the pointing error state of the current advancement time. The azimuth deviation and pitch deviation are sequentially determined to be within the safe domain boundary of the moving field of view. The advancement time when the azimuth deviation or pitch deviation first crosses the corresponding boundary is determined as the first exit time. The lower limit, upper limit, and interval probability of arrival time intervals are extracted sequentially from the valid message arrival distribution. The survival condition of the corresponding arrival time interval is determined as the first exit time not being less than the upper limit of the arrival time interval. The probability of the pointing error state remaining in the mobile field of view safe domain within the arrival time interval is calculated using the model transfer density to generate the survival probability. The survival probability is multiplied by the corresponding interval probability to obtain the interval survival contribution value. The interval survival contribution values ​​of all arrival time intervals are summed to generate the communication silent window survival function. Perform a natural logarithmic transformation on the communication silent window survival function, changing the azimuth deviation, pitch deviation, azimuth change rate, and pitch change rate in the pointing error state respectively. Calculate the ratio of the natural logarithmic transformation result to the change in each dimension component. Arrange the ratios corresponding to each dimension component in the order of the pointing error state to generate the state gradient of the communication silent window survival function. Perform matrix multiplication on the diffusion matrix and its transpose to generate the diffusion covariance matrix. Perform matrix multiplication on the diffusion covariance matrix and the state gradient of the communication silent window survival function to generate the field of view survival drift increment. Add the field of view survival drift increment to the reconstructed drift term to form the updated drift term.

[0029] In this embodiment, step eight specifically includes: Using the current pointing error state as the initial state, the updated drift term and diffusion matrix are substituted into the improved Ornstein-Uhlenbeck model. The time interval for each action completion is divided according to the target sampling time interval. At each target sampling time, the deterministic state increment is calculated based on the updated drift term, and the random state increment is calculated based on the diffusion matrix. The deterministic state increment and the random state increment are superimposed on the pointing error state corresponding to the previous target sampling time, and the predicted pointing error state sequence corresponding to the action completion time interval is generated sequentially. The target line-of-sight deviation is extracted from the predicted pointing error state sequence. The target line-of-sight deviation is decomposed into azimuth deviation and pitch deviation. The azimuth deviation is added to the gimbal azimuth angle to obtain the target line-of-sight azimuth angle. The pitch deviation is added to the gimbal pitch angle to obtain the target line-of-sight pitch angle. The target line-of-sight azimuth angle and the target line-of-sight pitch angle are combined according to the target sampling time to form a target line-of-sight angle sequence. Then, the target line-of-sight angle sequence corresponding to each action completion time interval is multiplied by the interval probability and weighted summation is performed to generate the target line-of-sight prediction trajectory. According to the target sampling time, the line-of-sight angles of adjacent targets are extracted from the target line-of-sight prediction trajectory. The difference between the target line-of-sight azimuth angle and the target line-of-sight pitch angle are calculated respectively. The difference between the sampling times of adjacent targets is also calculated. The azimuth rate is obtained by dividing the difference between the target line-of-sight azimuth angles by the difference between the sampling times of adjacent targets. The pitch rate is obtained by dividing the difference between the target line-of-sight pitch angles by the difference between the sampling times of adjacent targets. The target line-of-sight azimuth angle, target line-of-sight pitch angle, azimuth rate and pitch rate are combined to generate the gimbal control quantity. Input the gimbal control signal into the gimbal to control the azimuth angle to change towards the target's line of sight and to control the pitch angle to change towards the target's line of sight. Adjust the azimuth rotation process according to the azimuth angular velocity and the pitch rotation process according to the pitch angular velocity, so that the gimbal's line of sight changes along the predicted trajectory of the target's line of sight, driving the gimbal to track the target.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a countermeasure robot gimbal tracking task in a closed low-altitude target test range. The test target was a low-altitude flying target capable of moving along a preset path, with the target distance to the countermeasure robot maintained within a range of 80 to 220 meters. The target's horizontal angular velocity was 4° / s to 18° / s, and its pitch angular velocity was 2° / s to 10° / s. The countermeasure robot moved along a curved path at a speed of 0.6 m / s to 1.5 m / s, with a yaw angle variation range of -25° to 25°, a pitch angle variation range of -6° to 8°, and a roll angle variation range of -5° to 5°. The gimbal had a horizontal field of view of 58°, a vertical field of view of 34°, a maximum angular velocity of 60° / s, and a maximum angular acceleration of 120° / s². The signal reception power of the narrowband IoT link ranges from -112dBm to -86dBm, the number of message retransmissions ranges from 0 to 3, the buffer queuing time ranges from 35ms to 220ms, and the communication completion time between message sending and receiving ranges from 160ms to 780ms. Each test lasts for 30 minutes, and a total of 10 comparative tests were completed under the same path and the same target motion parameters.

[0031] During the test, the imaging device mounted on the gimbal acquired target images at a frequency of 25 frames per second. From each frame, the target center pixel coordinates and target bounding box scale were extracted, and the target sampling time was recorded. The narrowband IoT communication module simultaneously recorded the message transmission time, message reception time, message retransmission count, signal reception power, and buffer queuing time. The robot controller acquired the robot's heading angle, pitch angle, and roll angle, while the gimbal controller acquired the gimbal azimuth angle, pitch angle, angular velocity, and angular acceleration. Target observation data, narrowband IoT link data, robot posture, gimbal field of view parameters, and gimbal status were correlated according to the target sampling time to form a tracking dataset.

[0032] The tracking dataset is sorted according to the target sampling time. The horizontal and vertical pixel offsets of the target center pixel coordinates relative to the imaging center are converted into the target line-of-sight azimuth and target line-of-sight pitch angles. Then, the coordinate rotation is completed using the attitude rotation matrix formed by the robot's posture. The angle difference between the coordinate rotation result and the gimbal azimuth and pitch angles is calculated to obtain the target line-of-sight deviation. The rate of change is calculated based on the change in target line-of-sight deviation between consecutive target sampling times, forming the pointing error state. In the initial stage of testing, the mean regression center, drift matrix, and diffusion matrix are calculated using the continuously acquired pointing error states to construct an improved Ornstein-Uhlenbeck model. The communication completion time is added to the gimbal mechanical response time to form an action completion time sample. Then, the action completion time distribution is generated according to the message retransmission count and the signal received power range.

[0033] Based on the action completion time distribution and gimbal status, the azimuth reachable angle, pitch reachable angle, azimuth reachable velocity, and pitch reachable velocity are calculated, and the terminal acquisition domain and terminal acquisition distribution are generated. The model transition density is used to calculate the state transition probability of the current pointing error state transitioning to the pointing error state within the terminal acquisition domain. The state transition probability, state probability, and interval probability together generate the terminal acquisition potential function, and the state gradient of the terminal acquisition potential function is used to reconstruct the drift term. The valid message arrival time samples are probability-normalized to form the valid message arrival distribution. The target line-of-sight deviation and the predicted offset are added to construct the moving field-of-view safety domain. By calculating the first exit time of the pointing error state leaving the moving field-of-sight safety domain, a communication silent window survival function is generated, and the drift term is updated using the state gradient of the communication silent window survival function. The updated drift term and diffusion matrix are substituted into the improved Ornstein-Uhlenbeck model to generate the target line-of-sight predicted trajectory and gimbal control variables. The comparison method uses fixed communication delay compensation and linear extrapolation of the target line-of-sight angle to control the gimbal, completing the tracking task under the same test conditions. The test results are shown in Table 1. Table 1 Comparison of Gimbal Tracking Performance

[0034] As shown in Table 1, this invention incorporates narrowband IoT link latency, gimbal mechanical response capability, robot posture changes, and target line-of-sight motion state into the improved Ornstein-Uhlenbeck model. Compared with the fixed-latency linear prediction method, the average target line-of-sight deviation is reduced by 56.7%, the 95th percentile of the target line-of-sight deviation is reduced by 63.4%, the number of times the field of view boundary is crossed is reduced by 82.4%, the number of times the gimbal swings back and forth is reduced by 77.7%, the average tracking interruption time is shortened by 76.1%, and the continuous tracking rate is increased by 7.3 percentage points. This indicates that this invention can maintain the matching relationship between the target line-of-sight predicted trajectory and the gimbal execution capability under conditions of message retransmission, buffer queuing, and robot posture changes, reduce tracking interruptions caused by the target going out of the gimbal's field of view, and improve the stability, accuracy, and anti-communication latency capability of the servo gimbal automatic tracking.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for automatic tracking of a countermeasure robot using a servo-mounted gimbal based on narrowband Internet of Things, characterized in that, Includes the following steps: Step 1: Collect target observation data and narrowband IoT link data to obtain robot posture, gimbal field of view parameters and gimbal status, forming a tracking dataset; Step 2: Perform temporal registration and coordinate mapping on the tracking dataset to generate a pointing error state consisting of target line-of-sight deviation and rate of change; Step 3: Construct an improved Ornstein-Uhlenbeck model with the pointing error state, and generate the action completion time distribution from the tracking dataset; Step 4: Construct the terminal capture domain based on the action completion time distribution and gimbal status, and generate the terminal capture distribution; Step 5: Based on the action completion time distribution and terminal capture distribution, process the model transition density, generate the terminal capture potential function, and reconstruct the drift term from the state gradient of the terminal capture potential function; Step 6: Generate valid messages from the tracking dataset to reach the distribution, and construct a mobile field of view safety domain based on the pointing error status and PTZ field of view parameters; Step 7: Calculate the first exit time of the pointing error state when it leaves the safe domain of the moving field of view, generate the communication silent window survival function by combining the distribution of valid message arrivals, and update the drift term by the state gradient of the communication silent window survival function. Step 8: Solve the model according to the drift term, generate the target line-of-sight prediction trajectory and convert it into gimbal control quantity to drive the gimbal to track the target.

2. The automatic tracking method for countermeasure robots using a gimbal based on narrowband IoT according to claim 1, characterized in that, Step one specifically involves: Collect the target center pixel coordinates and target bounding box scale and record the target sampling time. Correlate the target center pixel coordinates and target bounding box scale with the target sampling time to form target observation data. Record the sending and receiving times of narrowband IoT messages, collect the number of message retransmissions, signal receiving power, and buffer queuing time to form narrowband IoT link data; Collect robot heading angle, robot pitch angle and robot roll angle to form robot posture, obtain gimbal horizontal field of view angle and gimbal vertical field of view angle to form gimbal field of view parameters, collect gimbal azimuth angle, gimbal pitch angle, gimbal angular velocity and gimbal angular acceleration to form gimbal state. Based on the target sampling time, message sending time, and message receiving time, target observation data and narrowband IoT link data are associated, and robot posture, gimbal field of view parameters, and gimbal status are bound to the target observation data to form a tracking dataset.

3. The automatic tracking method for countermeasure robots using a gimbal based on narrowband IoT according to claim 1, characterized in that, Step two specifically involves: The tracking dataset is sorted according to the target sampling time, and the target center pixel coordinates are associated with the corresponding robot posture and gimbal status to generate a time-series registration record. Calculate the horizontal and vertical pixel offsets of the target center pixel coordinates relative to the imaging center, and convert the horizontal and vertical pixel offsets into the target line-of-sight azimuth and target line-of-sight pitch angles according to the gimbal field of view parameters to generate the target line-of-sight angle; The robot's posture is converted into a posture rotation matrix. The posture rotation matrix is ​​used to perform coordinate rotation on the target line of sight angle. The coordinate rotation result is compared with the gimbal azimuth and gimbal pitch angles in the gimbal state to calculate the angle difference and generate the target line of sight deviation. Calculate the difference in target line-of-sight deviation and the difference in target sampling time corresponding to consecutive target sampling times. Divide the difference in target line-of-sight deviation by the difference in target sampling time to generate the rate of change. Combine the target line-of-sight deviation and the rate of change according to the target sampling time to generate the pointing error state.

4. The automatic tracking method for countermeasure robots using a gimbal based on narrowband IoT according to claim 1, characterized in that, Step three specifically involves: Arrange the pointing error states according to the target sampling time, combine the target line-of-sight deviation and the rate of change into a state vector sequence, and calculate the mean regression center and the increment of adjacent state vectors of the state vector sequence. Using the deviation of the state vector from the mean regression center as the independent variable and the increment of the adjacent state vector as the dependent variable, least squares fitting is performed to generate the drift matrix. The fitting error covariance is calculated and decomposed to generate the diffusion matrix. The mean regression center, drift matrix, and diffusion matrix are written into the Ornstein-Uhlenbeck state equation to construct the improved Ornstein-Uhlenbeck model and generate the model transition density. Extract the target sampling time, message reception time, message retransmission count, signal reception power, and PTZ status from the tracking dataset. Subtract the target sampling time from the message reception time to generate the communication completion time. Calculate the PTZ mechanical response time when the PTZ angular displacement reaches the target line-of-sight deviation based on the target line-of-sight deviation and PTZ status. The action completion time sample is generated by adding the communication completion time and the gimbal mechanical response time. The action completion time sample is divided according to the message retransmission count and the signal received power range. The frequency of occurrence of the action completion time sample in each range is counted and the probability is normalized to generate the action completion time distribution.

5. The automatic tracking method for countermeasure robots using a gimbal based on narrowband IoT according to claim 1, characterized in that, Step four specifically involves: Extract the action completion time interval and interval probability from the action completion time distribution, and use the gimbal azimuth angle, gimbal pitch angle, gimbal angular velocity and gimbal angular acceleration to calculate the azimuth reachable angle, pitch reachable angle, azimuth reachable velocity and pitch reachable velocity corresponding to the action completion time interval. The target line-of-sight deviation is compared with the azimuth reachable angle and the pitch reachable angle, and the rate of change is compared with the azimuth reachable speed and the pitch reachable speed. Pointing error states that meet the angle reachable condition and the speed reachable condition are selected to generate the capture subdomain. The capture subdomains are merged according to the action completion time interval, and the interval probability is configured to the corresponding capture subdomain to generate a terminal capture domain consisting of the target line-of-sight deviation range, the change speed range, and the interval probability. Calculate the target line-of-sight deviation amplitude and change rate amplitude corresponding to each pointing error state within the terminal capture domain. Generate state weights based on the target line-of-sight deviation amplitude and change rate amplitude. Multiply the state weights by the interval probability and perform probability normalization to generate the terminal capture distribution.

6. The automatic tracking method for a countermeasure robot servo gimbal based on narrowband Internet of Things according to claim 1, characterized in that, Step five specifically involves: Extract the action completion time interval and interval probability from the action completion time distribution, and extract the state probability corresponding to the pointing error state in the terminal capture domain from the terminal capture distribution; Using the current pointing error state as the starting state and the pointing error state in the terminal capture domain as the ending state, the action completion time interval is substituted into the model transition density to calculate the state transition probability from the starting state to the ending state. Multiply the state transition probability, state probability, and interval probability, sum them according to the terminal capture domain and the action completion time interval to generate the terminal capture potential function. Calculate the partial derivative of the terminal capture potential function with respect to each dimension of the pointing error state to generate the state gradient of the terminal capture potential function. Multiply the diffusion matrix by its transpose, then multiply the product matrix by the gradient of the terminal capture potential function state to generate the terminal capture drift increment. This terminal capture drift increment is then added to the drift matrix and applied to the base drift amount formed by the difference between the pointing error state and the mean regression center, thus completing the drift term reconstruction.

7. The automatic tracking method for a countermeasure robot servo gimbal based on narrowband IoT according to claim 1, characterized in that, The model transition density takes the initial pointing error state, the final pointing error state, and the state transition duration as inputs. It calculates the state transition mean based on the mean regression center, the drift matrix, and the initial pointing error state. It calculates the state transition covariance based on the drift matrix, the diffusion matrix, and the state transition duration. It calculates the probability density based on the state deviation between the final pointing error state and the state transition mean, as well as the state transition covariance. This density is used to characterize the probability density of the initial pointing error state transitioning to the final pointing error state within the state transition duration.

8. The automatic tracking method for countermeasure robots using a gimbal based on narrowband IoT according to claim 1, characterized in that, Step six specifically involves: Extract the target sampling time and message reception time from the tracking dataset, perform time-series matching between the message reception time and the target sampling time, retain the messages that are associated with the target observation data, and generate valid message records; Arrange valid message records according to message reception time, calculate the difference between reception times of adjacent messages, generate valid message arrival time samples, divide valid message arrival time samples according to message retransmission count and signal reception power range, count the occurrence frequency of valid message arrival time samples and perform probability normalization to generate valid message arrival distribution. Extract the target line-of-sight deviation and rate of change from the pointing error state, and determine half of the gimbal's horizontal field of view as the azimuth field of view boundary and half of the gimbal's vertical field of view as the pitch field of view boundary. The arrival time intervals in the valid message arrival distribution are multiplied by the rate of change to generate a predicted offset. The predicted offset is added to the target line-of-sight deviation. The pointing error state of the summation result within the azimuth and pitch field-of-sight boundaries is retained. The pointing error states are then merged according to the arrival time intervals and interval probabilities to generate a moving field-of-sight safety domain.

9. The automatic tracking method for a countermeasure robot servo gimbal based on narrowband Internet of Things according to claim 1, characterized in that, Step seven specifically involves: Using the current pointing error state as the initial state, the pointing error state is advanced by the reconstructed drift term and diffusion matrix of the improved Ornstein-Uhlenbeck model. The pointing error state at each advancement moment is compared with the boundary of the moving field of view safety domain. The advancement moment when the pointing error state first crosses the boundary of the moving field of view safety domain is determined as the first exit time. The arrival time interval and interval probability are extracted from the valid message arrival distribution. The survival condition is that the first exit time is not less than the upper limit of the arrival time interval. The survival probability corresponding to the survival condition is calculated based on the model transition density. The survival probability is multiplied by the interval probability and summed according to the arrival time interval to generate the communication silent window survival function. Perform a natural logarithmic transformation on the communication silent window survival function, calculate the partial derivatives of the natural logarithmic transformation result with respect to each dimension of the pointing error state, and generate the state gradient of the communication silent window survival function. Multiply the diffusion matrix by its transpose, then multiply the product matrix by the state gradient of the communication silent window survival function to generate the field-of-view survival drift increment. Finally, add the field-of-view survival drift increment to the reconstructed drift term to complete the drift term update.

10. The automatic tracking method for a countermeasure robot servo gimbal based on narrowband Internet of Things according to claim 1, characterized in that, Step eight specifically involves: Using the current pointing error state as the initial state, the updated drift term and diffusion matrix are substituted into the improved Ornstein-Uhlenbeck model. The pointing error state is advanced within each action completion time interval according to the target sampling time interval, generating the predicted pointing error state sequence corresponding to the action completion time interval. The target line-of-sight deviation is extracted from the predicted pointing error state sequence. The target line-of-sight deviation is then superimposed with the gimbal azimuth and gimbal pitch angles to generate a target line-of-sight angle sequence corresponding to the action completion time interval. The target line-of-sight angle sequence is weighted and summarized using interval probability to generate the target line-of-sight prediction trajectory. Calculate the angle difference between adjacent target line-of-sight angles and the difference in target sampling time in the target line-of-sight prediction trajectory. Divide the angle difference by the difference in target sampling time to generate the gimbal angular velocity. Combine the target line-of-sight angle and gimbal angular velocity in the target line-of-sight prediction trajectory to generate the gimbal control quantity. Input the gimbal control parameters into the gimbal to control the azimuth and pitch angles of the gimbal to change according to the predicted trajectory of the target's line of sight, and adjust the rotation process according to the angular velocity of the gimbal to drive the gimbal to track the target.

Images