Unmanned aerial vehicle maneuvering target prediction and tracking method based on improved dynamic window method

Abstract

The invention provides an unmanned aerial vehicle maneuvering target prediction and tracking method based on an improved dynamic window method, and the method comprises the following steps: S1, building an unmanned aerial vehicle kinematic model and maneuvering target multi-model state estimation, and obtaining a constant speed CV model, a constant steering rate speed CTRV model and an unmanned aerial vehicle system state equation of a maneuvering target; s2, based on the model, adopting an IMM-EKF algorithm of a self-adaptive model probability to obtain a prediction track cluster of the maneuvering target; s3, constructing a potential field virtual target point based on the predicted trajectory cluster of the maneuvering target and an unmanned aerial vehicle system state equation; and S4, combining the potential field virtual target point with the improved DWA algorithm to obtain an unmanned aerial vehicle control instruction, and executing the instruction to perform target prediction and tracking. Compared with the prior art, the method has the advantages that the technical problems that a traditional DWA algorithm is insufficient in target maneuverability adaptation, prone to falling into local optimum and low in tracking efficiency in maneuvering target tracking are solved, and the like.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and decision-making technology for unmanned aerial vehicles (UAVs), and in particular to a method for predicting and tracking maneuvering targets for UAVs based on an improved dynamic window method. Background Technology

[0002] Unmanned aerial vehicle (UAV) target tracking technology has significant application value in fields such as security inspection, search and rescue, and logistics delivery. The traditional Dynamic Window (DWA) method, as an effective local path planning algorithm, is widely used in UAV target tracking. However, the traditional DWA algorithm suffers from the following problems in tracking maneuvering targets:

[0003] (1) Insufficient adaptation to target mobility: Traditional DWA algorithm mainly makes tracking decisions based on the current target position, lacks the ability to predict the future movement trend of the target, and is difficult to effectively deal with the sudden turning or acceleration behavior of the target.

[0004] (2) Easily trapped in local optima: In complex obstacle environments, the DWA algorithm is prone to generating “follow-type” tracking trajectories, causing the UAV to passively follow the target’s historical trajectory and lack the ability to make proactive decisions.

[0005] (3) Low tracking efficiency: The evaluation function of the traditional DWA algorithm is relatively simple, making it difficult to ensure both tracking accuracy and flight efficiency and safety.

[0006] In the existing technology, although some researchers have proposed improved schemes based on artificial potential field method, reinforcement learning and other methods, these methods have high computational complexity or lack in-depth consideration of the motion characteristics of maneuvering targets, resulting in low efficiency of target prediction and tracking, and insufficient flexibility and adaptability. Summary of the Invention

[0007] The purpose of this invention is to provide a UAV maneuvering target prediction and tracking method based on the improved dynamic window method, which solves the technical problems of the traditional DWA algorithm in maneuvering target tracking, such as insufficient adaptability to target maneuverability, easy getting trapped in local optima, and low tracking efficiency.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] Compared with the prior art, the present invention has the following beneficial effects:

[0010] This invention significantly improves the efficiency of UAVs in recognizing and switching between different motion modes of maneuvering targets by employing the IMM-EKF framework and adaptive model probabilistic optimization, effectively solving the problem of insufficient adaptability to target maneuverability in traditional methods. Furthermore, this invention utilizes a potential field virtual target method to achieve safe and flexible obstacle avoidance and adaptive active tracking in complex obstacle environments, effectively reducing the risk of the algorithm getting trapped in local optima. Finally, this invention significantly improves tracking efficiency by improving the DWA evaluation function and weighted recursive strategy.

[0011] This invention utilizes the IMM-EKF framework combined with an adaptive model probability optimization mechanism to dynamically adjust the transition probability weights of the CV and CTRV models based on real-time environmental information such as the relative distance between the maneuvering target and obstacles, pitch angle, and yaw angle. This enables the system to quickly identify changes in target motion patterns and switch to the corresponding motion model in a timely manner, significantly improving the efficiency of UAV in recognizing and switching between different motion patterns of maneuvering targets. By constructing a dynamic coupling mechanism between a dual attractive potential field model and an obstacle repulsive potential field, combined with a benchmark virtual target point selection strategy for real-time collision detection, the UAV can adaptively switch between traditional follow mode and active interception mode, overcoming the local optima problem caused by the single target point in traditional DWA algorithms and effectively reducing the risk of the algorithm getting trapped in local optima. Finally, the single target distance term in the traditional DWA evaluation function is refined into three independent evaluation dimensions: maneuvering target state estimation distance term, maneuvering target state prediction distance term, and potential field virtual target point distance term. A new altitude safety distance term is added, and a weighted recursive strategy is used to smooth control commands, enabling the UAV to simultaneously consider current tracking accuracy, future motion prediction, and environmental safety, significantly improving tracking efficiency.

[0012] While maintaining high linear velocity stability, this invention also maintains relatively stable performance in terms of angular velocity, exhibiting more stable and smoother characteristics than traditional algorithms, demonstrating greater flexibility and adaptability. By fusing multimodal information, it achieves proactive prediction-interception tracking, breaking away from the limitations of the "follow-the-path" mode of traditional tracking algorithms. By predicting the target's motion trend in advance, it completes lateral interception tracking, significantly improving tracking efficiency. Attached Figure Description

[0013] Figure 1 This is a schematic diagram illustrating the principle of constructing virtual target points in the potential field during maneuvering target tracking in this invention.

[0014] Figure 2 This is a comparison chart of experimental results for multiple methods of UAV target trajectory tracking and prediction according to the present invention;

[0015] Figure 3 This is an experimental scenario diagram of the UAV target trajectory tracking and prediction method of the present invention;

[0016] Figure 4 This is a system flowchart of the UAV target trajectory tracking and prediction method of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0018] This invention proposes a method for predicting and tracking maneuvering targets in unmanned aerial vehicles (UAVs) based on an improved dynamic window method, comprising the following steps: S1, establishing a UAV kinematic model and a multi-model state estimation of the maneuvering target; S2, designing a maneuvering target prediction strategy based on an Interacting Multiple Model-Extended Kalman Filter (IMM-EKF) framework; S3, constructing a potential field virtual target point generation mechanism; S4, improving the evaluation function of the traditional Dynamic Window Approach (DWA) to achieve maneuvering target tracking. This invention, based on a fusion framework of the maneuvering target prediction strategy and the potential field virtual target method, effectively solves the problems of traditional dynamic window methods easily getting trapped in local optima and having low tracking efficiency in UAV maneuvering target tracking, significantly improving the maneuvering target tracking performance of UAVs in complex obstacle environments.

[0019] The steps are as follows:

[0020] S1. Establish the kinematic model of the UAV and the multi-model state estimation of the maneuvering target;

[0021] S2. Design a maneuvering target prediction strategy based on the IMM-EKF framework;

[0022] S3. Construct a mechanism for generating virtual target points in the potential field;

[0023] S4. Improve the DWA algorithm evaluation function to achieve maneuvering target tracking.

[0024] In the traditional IMM-EKF algorithm, the transition matrix remains unchanged throughout the process and cannot be actively adjusted according to environmental changes. This invention adds an environment-aware adaptive adjustment mechanism. Based on the current environmental perception trigger, the transition probabilities are actively adjusted (reducing the weights of the CV model and increasing the weights of the CTRV model), and then the standard IMM-EKF algorithm is executed, using the adjusted transition probability matrix for Bayesian updates.

[0025] Step S1 further includes the following steps:

[0026] S11. Establish a three-dimensional spatial kinematic model of the UAV. Considering the motion characteristics of the UAV target in three-dimensional space, establish a discrete-time domain kinematic model, defining the state vector at time k as...

[0027]

[0028] Where, x k ,y k ,z k Let θ be the position of the UAV in the three-dimensional coordinate system. k and ψ k These are the pitch and yaw angles of the drone, respectively. k , and These correspond to the linear velocity, pitch rate, and yaw rate of the UAV, respectively.

[0029] S12. Establish a constant speed CV model and a constant turning rate speed CTRV model for the maneuvering target. Based on the motion characteristics of the maneuvering target, select the CV model and CTRV model to construct a multi-model estimation framework.

[0030] S13. Establish the system state equation X k+1 =F·X k +W k Where F is the system state transition matrix, W k This refers to system process noise.

[0031] The state transition matrix F of the CV model CV Designed as follows:

[0032]

[0033] in:

[0034]

[0035] Where I n Represents an n-dimensional identity matrix, 0 m×n For an m×n zero matrix, θ k and ψ k These are the pitch and yaw angles of the drone, respectively. k Let T be the linear velocity, and T represent the sampling time interval.

[0036] The CTRV model derives the position update equation through analytical integration:

[0037]

[0038] In the formula, x k ,y k ,z k Let v be the current position coordinates of the UAV at time k.k Let θ represent the linear velocity of the UAV at time k. k Let ψ represent the pitch angle of the UAV at time k. k This represents the yaw angle of the UAV at time k. This represents the pitch angular velocity of the UAV at time k. Let represent the yaw rate of the UAV at time k, and T represent the sampling time interval. To improve computational efficiency, an approximation scheme based on Taylor expansion is adopted:

[0039]

[0040] In the formula e l F is a standard basis vector with the l-th element being 1. CV This is the state transition matrix of the CV model.

[0041] Step S2 further includes the following steps:

[0042] S21. Construct an IMM-EKF multi-model interactive architecture to realize the estimation and prediction of the motion state of maneuvering targets. Its core consists of three parts: a multi-model hybrid filtering module, a Bayesian probability evaluation module, and an optimal estimation fusion module.

[0043] S22. An adaptive model probability optimization mechanism is designed. The traditional IMM-EKF algorithm uses fixed transition probabilities, which are insufficient to effectively handle sudden turning or obstacle avoidance behaviors of maneuvering targets. Therefore, an adaptive transition probability optimization mechanism based on environmental perception is proposed. By dynamically adjusting the transition probability weights between models, the algorithm's real-time response capability to target maneuver mode switching is enhanced. Detailed mathematical implementation of the IMM-EKF algorithm:

[0044] The model transition probabilities were calculated using normalization to ensure the validity of the probability distribution. The initial conditions for matching each motion model to the EKF were obtained through parameter calculation using a Gaussian mixture distribution, ensuring the mathematical rigor of multi-model fusion.

[0045] The likelihood function is calculated based on the measurement residuals and covariance matrix, and is implemented through the probability density function of the Gaussian distribution, providing a theoretical basis for updating the model weights.

[0046] S23. Based on the state estimate at time k and the optimized model probability, the algorithm generates a prediction sequence for the next N steps through forward recursion.

[0047] The specific implementation of the IMM-EKF algorithm includes four steps:

[0048] (1) Calculation of model transition probabilities;

[0049] (2) Calculate the initial conditions for matching EKF for each motion model;

[0050] (3) Using the initial conditions, perform EKF iterations for matching each model;

[0051] (4) Obtain the final state estimate and covariance matrix.

[0052] S24. Calculate the transition probability using the model:

[0053]

[0054] In the formula, It is the model transition probability from model i to j at time k-1. The weight coefficients of model i at time k-1 are... Here, r is the normalization factor, and r is the total number of models.

[0055] S25. Calculate the likelihood function:

[0056]

[0057] In the formula, Let be the likelihood function of the j-th motion model at time k. The measurement residual of the j-th model, The residual covariance matrix of the j-th model, where r is the total number of models.

[0058] S26. Update model probabilities:

[0059]

[0060] In the formula, Let c be the likelihood function of the j-th motion model at time k. j This represents the mixture probability of the j-th model. Let r be the update probability of the j-th model at time k, and r be the total number of models.

[0061] The adaptive model probability optimization mechanism is as follows:

[0062] When condition d is satisfied k <d thres 、|θ k |>θ thres or |ψ k |>ψ thres At this time, the execution probability is adjusted:

[0063]

[0064] Otherwise, perform the reverse adjustment:

[0065]

[0066] in, Let be the transition probability of the CV model at time k-1. Let be the transition probability of the CTRV model at time k-1. Where δ∈(0,0.1] is the probability adjustment step size, and μ... min =0.005 is the lower limit of the model weights; the adjusted transition probability matrix is ​​normalized to ensure that the sum of the probabilities in each row is 1.

[0067] The implementation steps of the adaptive transition probability optimization mechanism are as follows:

[0068] (1) Maneuver Feature Extraction: Real-time acquisition of the relative distance d between the maneuvering target and obstacles. k Pitch angle θ of maneuvering target k and the yaw angle χ of the maneuvering target k .

[0069] (2) Setting threshold parameters: Based on the kinematic parameters of the maneuvering target and the statistical distribution of the data, set the safe distance threshold d. thres Pitch angle threshold θ thres and yaw angle threshold χ thres .

[0070] (3) Dynamic adjustment of transition probability: when condition d is met k <d thres 、|θ k |>θ thres or |ψ k |>ψ thres At one time;

[0071] Trigger probability adjustment:

[0072] Otherwise, perform the reverse adjustment:

[0073] in, Let be the transition probability of the CV model at time k-1. Let be the transition probability of the CTRV model at time k-1. Where δ∈(0,0.1] is the probability adjustment step size, and μ... min =0.005 is the lower limit of the model weights; the adjusted transition probability matrix is ​​normalized to ensure that the sum of the probabilities in each row is 1.

[0074] Based on the state estimate at time k and the optimized model probability, the algorithm generates a prediction sequence for the next N steps through forward recursion.

[0075] The system measurement model is represented by Z. k =HX k +V k Z k It is the measurement vector, H = I8 is the observation matrix, V k For measuring noise.

[0076] Obtain the final state estimate and covariance matrix:

[0077] Final state estimation

[0078] in, For fusion state estimation, The probability weights of the j-th model, This is the state estimate for the j-th model.

[0079] covariance matrix

[0080] In the formula, P k|k To integrate the covariance matrix, Let be the covariance matrix of the j-th model. The probability weights of the j-th model, For fusion state estimation, This is the state estimate for the j-th model.

[0081] The multi-step prediction process fully considers the probability weights of each model and generates a cluster of probabilistic prediction trajectories through iterative kinematic equations.

[0082] This prediction paradigm provides key inputs for the potential field virtual target method and the DWA motion planning algorithm, thereby driving the UAV to perform efficient maneuvering target tracking.

[0083] S3 further includes the following steps:

[0084] S31. Construct a dual attraction potential field model, including: an instantaneous attraction potential field, generated from the estimated position of the maneuvering target at the current moment, which guides the virtual target point to directly track; and a predicted attraction potential field, generated from the predicted position of the maneuvering target in the next N steps, which guides the virtual target point to move in the direction of the target's movement trend.

[0085] S32, Based on the current location X of the drone k and estimated position of maneuvering targets Generating traditional virtual target points

[0086] S33, Based on the current location X of the drone k and predicted location of maneuvering targets Generate predicted virtual target points

[0087] S34. Select a baseline virtual target point through real-time collision detection;

[0088] S35. Use the potential field force to correct the reference virtual target point and generate the potential field virtual target point;

[0089] S36. Establish a dynamic coupling mechanism between the dual attractive potential field and the obstacle repulsive potential field to achieve autonomous optimization of multimodal trajectories; the principle of constructing virtual target points in the potential field is based on the dynamic coupling between the dual attractive potential field and the obstacle repulsive potential field to achieve real-time optimization of the tracking trajectory. The construction of virtual target points in the potential field is based on the attractive force F generated by the two target points. att1 With F att2 The repulsive force F generated by the obstacle rep Dynamic coupling generates a virtual target point in the potential field through combined force, enabling autonomous selection of multimodal trajectories.

[0090] The formula for calculating the predicted virtual target point is:

[0091] In the formula, k represents the current time k. Predict the virtual target point, X k Indicates the current location of the drone. The predicted position of the maneuvering target is N steps, where λ is the distance parameter to the virtual target point.

[0092] Generation of virtual target points in potential field:

[0093] In the formula, k represents the current time k. Virtual target point of potential field Let λ be the reference virtual target point, λ represent the distance correction parameter, and F be the resultant force vector.

[0094] Baseline virtual target point selection mechanism:

[0095] Where k represents the current time k. As a reference virtual target point, Represents a traditional virtual target point. To predict virtual target points, the potential field virtual target point is generated as follows:

[0096] Guide the drone to judge and select different feasible trajectories.

[0097] This dual gravitational field cooperative mechanism embeds forward-looking interception requirements while maintaining target tracking. Through adaptive fusion of field strength vectors, it enables UAVs to build a continuous decision space between passive following and active interception modes, thereby simultaneously improving real-time obstacle avoidance robustness and maneuvering target tracking efficiency.

[0098] S37, Building Attraction F att1 and F att2 repulsive force F from the obstacle rep A potential field virtual target point is dynamically generated through coupling. For two significantly different candidate paths (exhibiting 0-1 decision-making characteristics), a baseline virtual target point selection mechanism is established to ensure the reliability of the trajectory decision.

[0099] The traditional formula for calculating virtual target points is:

[0100]

[0101] in, For traditional virtual target points, X k Indicates the current location of the drone. Here, λ is the estimated position of the maneuvering target state, and λ is the distance parameter of the virtual target point. The formula for calculating the predicted virtual target point is:

[0102]

[0103] In the formula, Predict the virtual target point, X k The drone's current location The predicted position of the maneuvering target in N steps, where λ is the distance parameter to the virtual target point; the mechanism for selecting the baseline virtual target point:

[0104]

[0105] in, As a reference virtual target point, Represents a traditional virtual target point. To predict virtual target points, the potential field virtual target point is generated as follows:

[0106]

[0107] In the formula, Virtual target point of potential field Let λ be the reference virtual target point, λ represent the distance correction parameter, and F be the resultant force vector.

[0108] The real-time collision detection module determines the drone's current position X. k Estimated position of maneuvering target The dynamic relationship between the motion points is dynamically evaluated, and traditional virtual target points or predicted virtual target points are autonomously selected as reference points based on the collision risk assessment results.

[0109] This selection mechanism ensures the reliability and safety of trajectory decisions in complex environments.

[0110] In step S4, the improved DWA evaluation function is:

[0111]

[0112] In the formula, α, β, γ, ∈, ζ, η, ξ, ρ are the weight coefficients of each indicator item; the weight coefficients of the evaluation function are dynamically switched according to the collision detection results to adjust the route planning and ensure smooth tracking. Among them, the pitch angle evaluation item... Yaw angle evaluation item Obstacle avoidance distance evaluation item and speed evaluation item vel k,s Consistent with traditional algorithms, the target distance evaluation term in the traditional evaluation function is refined into a maneuvering target state estimation distance term in the improved evaluation function. Predicted range term for maneuvering target status Distance term of virtual target point in potential field To reduce the risks associated with drones flying too low, a height safety distance parameter was introduced.

[0113] The formula for calculating the height safety distance is:

[0114]

[0115] In the formula, The high security score for the s-th candidate trajectory at time k. d represents the height at the end of the trajectory. safe Safety altitude threshold. The formula for calculating the distance term in the maneuvering target state estimation is:

[0116]

[0117] In the formula, Let S be the distance score between the s-th candidate trajectory at time k and the estimated position of the maneuvering target. Let be the distance between the s-th trajectory and the estimated position of the target. This represents the sum of the distances for all trajectories. The rest... The calculation methods for the two items are similar, so they will not be repeated here.

[0118] Weighted recursion strategy smooths linear velocity and attitude angular velocity commands:

[0119]

[0120] Where χ∈(0,1) is the instruction weight coefficient. This represents the smoothed control command, u k / k-1 The original control command at time k / k-1.

[0121] In step S4, the evaluation function of the DWA algorithm is improved by performing multi-objective collaborative optimization on the traditional DWA evaluation function based on the maneuvering target prediction strategy and the potential field virtual target method.

[0122] The basic principle of the DWA algorithm is based on the UAV kinematic model and dynamic constraints, constructing the UAV velocity sampling space V. r It is constrained by the maximum speed V s Reachable speed constraint V a and safety distance constraint V d The intersection of the three constitutes the whole.

[0123] The implementation steps of the adaptive model probability optimization mechanism are as follows:

[0124] (1) Maneuver Feature Extraction: Real-time acquisition of the relative distance d between the maneuvering target and obstacles. k Pitch angle θ of maneuvering target k and the yaw angle ψ of the maneuvering target k ;

[0125] (2) Setting threshold parameters: Based on the kinematic parameters of the maneuvering target and the statistical distribution of the data, set the safe distance threshold d. thres Pitch angle threshold θ thres and yaw angle threshold ψ thres ;

[0126] (3) Dynamic adjustment of transition probability: when condition d is met k <d thres 、|θ k |>θ thres or |ψ k |>ψ thres When one of these conditions is met, the probability is adjusted.

[0127] (4) Probability normalization constraint: The adjusted transition probability matrix is ​​normalized to ensure that the sum of probabilities in each row is 1;

[0128] (5) Adaptive transfer probability optimization mechanism: By introducing an adaptive model probability update mechanism, the response speed of the traditional IMM-EKF algorithm to sudden changes in the motion mode of the maneuvering target (such as emergency turning or sudden acceleration) is significantly improved.

[0129] The platform constructs a 3D simulation scene. The test environment is a 100m×100m×30m 3D grid space, which includes 3D terrain and randomly distributed cylindrical obstacles. The starting point of the maneuvering target is (10,54,20), the ending point is (60,55,22), and the initial position of the UAV is (13,70,20).

[0130] Simulation results show that, compared with the traditional APF algorithm, DWA algorithm, and VT-DWA algorithm, the MTP-PVT-DWA algorithm of this invention has significant improvements in path length, number of planning steps, and time consumption.

[0131] The path length is reduced by 44.8% compared to the VT-DWA algorithm;

[0132] The number of planned steps and the time spent were reduced by 37.1% and 37.2%, respectively;

[0133] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0134] Since the real-time collision detection results during the construction of the potential field virtual target point directly affect the selection of the tracking route, the weight coefficients of the evaluation function will be switched accordingly to adjust the route planning and ensure the smooth progress of tracking.

[0135] This includes employing a weighted recursive strategy to smooth linear velocity and attitude angular velocity commands, further improving the reliability of UAVs in tracking maneuvering targets in complex obstacle environments;

[0136] The performance metrics of the method include path length, number of planning steps, computation time, and tracking accuracy. Comparative experiments with traditional APF, DWA, and VT-DWA algorithms have verified the superiority of the method of the present invention.

[0137] It is applied to the tracking of maneuvering targets by UAVs in complex obstacle environments. It achieves active prediction-interception tracking through multimodal information fusion. The path length of the traditional DWA algorithm in the project is reduced by 44.8%, and the number of planning steps and time are reduced by 37.1% and 37.2%, respectively.

[0138] The beneficial effects of this invention are as follows:

[0139] (1) By using the IMM-EKF framework and adaptive model probability optimization, the efficiency of UAV in recognizing and switching different motion modes of maneuvering targets is significantly improved, effectively solving the problem of insufficient adaptation of traditional methods to target maneuverability;

[0140] (2) By using the potential field virtual target method, safe and flexible obstacle avoidance and adaptive active tracking are achieved in complex obstacle environments, which effectively reduces the risk of the algorithm getting stuck in local optima;

[0141] (3) By improving the DWA evaluation function and weighted recursion strategy, the tracking efficiency was significantly improved. Compared with the traditional DWA algorithm, the path length was reduced by 44.8%, and the number of planning steps and time consumption were reduced by 37.1% and 37.2%, respectively.

[0142] (4) The algorithm has good engineering applicability and real-time performance, and can run stably in complex dynamic environments.

[0143] (5) Active prediction-interception tracking is achieved through multimodal information fusion, which breaks away from the limitations of the traditional tracking algorithm's "following" path mode. By predicting the target's movement trend in advance, lateral interception tracking is completed, which greatly improves tracking efficiency.

[0144] (6) While maintaining high linear velocity stability, the algorithm also maintains relatively stable performance in terms of angular velocity. Overall, it exhibits more stable and smoother characteristics than traditional algorithms, demonstrating greater flexibility and adaptability.

[0145] In this invention, considering the motion characteristics of the UAV target in three-dimensional space, a discrete-time kinematic model is established (sampling period T). The state vector at time k is defined, and its equation of motion can be expressed as:

[0146]

[0147] Where, x k+1 y k+1 , z k+1 Let θ(t) be the coordinate position at time k+1, ψ(t) be the pitch angle and yaw angle of the UAV at time k+1, and v(t) be the current linear velocity. Let X be the pitch angular velocity and yaw angular velocity at time k. Based on the above kinematic model, the system state equation is established: X k+1 =F·X k +W k .

[0148] Figure 1 This is a schematic diagram illustrating the principle of constructing virtual target points in the potential field during maneuvering target tracking in this invention. Figure 2 This is a comparison chart of experimental results for multiple methods of UAV target trajectory tracking and prediction according to the present invention; Figure 3 This is an experimental scenario diagram of the UAV target trajectory tracking and prediction method of the present invention; Figure 4 This is a system flowchart of the UAV target trajectory tracking and prediction method of the present invention.

[0149] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for predicting and tracking maneuvering targets using an unmanned aerial vehicle (UAV) based on an improved dynamic window method, characterized in that, The method includes the following steps: S1. Establish the kinematic model of the UAV and the multi-model state estimation of the maneuvering target to obtain the constant velocity CV model, constant turning rate velocity CTRV model and the state equation of the UAV system. S2. Based on the model, the IMM-EKF algorithm with adaptive model probability is used to obtain the predicted trajectory cluster of the maneuvering target; S3. Construct virtual target points in the potential field based on the predicted trajectory clusters of maneuvering targets and the state equations of the UAV system; S4. Combine the potential field virtual target point with the improved DWA algorithm to obtain UAV control commands, and execute the commands to predict and track targets.

2. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 1, characterized in that, The specific steps of S2 are as follows: S21. Initialize the model parameters and transition probabilities to obtain the current transition probabilities and current model parameters; S22. Based on the initial current transition probability and the current model parameters, perform EKF updates and iterations for each model matching; S23. Calculate the likelihood function, update the model probability, and obtain the fusion state estimate; S24. Adjust the current transition probability based on the adaptive model probability mechanism; S25. Update the current time, use the adjusted transition probability as the current transition probability, and return to S22 to predict the trajectory cluster of the maneuvering target for the next N steps.

3. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 2, characterized in that, The specific steps for adjusting the current transition probability based on the adaptive model probability mechanism are as follows: Real-time acquisition of the relative distance d between the moving target and the obstacle k Pitch angle θ of maneuvering target k and the yaw angle ψ of the maneuvering target k ; Based on the kinematic parameters and statistical distribution of the maneuvering target, a safe distance threshold d is set. thres Pitch angle threshold θ thres and yaw angle threshold ψ thres ; Determine if d is satisfied k <d thres or |θ k |>θ thres or |ψ k |>ψ thres If so, then the transition probability is adjusted as follows: Conversely, the adjusted transition probability is: in, The transition probabilities after adjustment for the CV model. Let μ be the transition probability adjusted by the CTRV model, δ∈(0,0.1] be the probability adjustment step size, and μ be the transition probability. min =0.005 is the lower limit of the model weights. The adjusted transition probability matrix is ​​normalized to ensure that the sum of the probabilities in each row is 1.

4. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 1, characterized in that, S3 includes the following steps: S31. Estimate the position based on the current position of the UAV and the first maneuvering target in the predicted trajectory cluster. Generating traditional virtual target points S32. Estimate the position based on the current position of the UAV and the Nth maneuvering target in the predicted trajectory cluster. Generate predicted virtual target points S33. Select a baseline virtual target point through real-time collision detection; S34. Determine the potential field virtual target point based on the benchmark virtual target point.

5. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 4, characterized in that, Traditional virtual target points for: Among them, X k λ represents the current position of the drone, and λ is the distance parameter to the virtual target point.

6. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 5, characterized in that, Predict virtual target point for:

7. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 6, characterized in that, The baseline virtual target point is: Here, `collision_detect` represents the collision detection result, when the drone's current position is close to the estimated position of the moving target. The value is true if there is an obstacle on the straight path between them, and false otherwise.

8. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 7, characterized in that, The virtual target point of the potential field is: in, Let F represent the virtual target point in the potential field, and F represent the resultant force vector. The calculation formula is: f=F att1 +F att2 +F rep Among them, F att1 Traditional virtual target point The attractive force F exerted on the virtual target point in the potential field att2 To predict virtual target points The attractive force F exerted on the virtual target point in the potential field rep This refers to the repulsive force exerted by environmental obstacles on the virtual target point in the potential field.

9. The method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on the improved dynamic window method according to claim 1, characterized in that, The specific steps for obtaining UAV control commands by combining the potential field virtual target point and the improved DWA algorithm are as follows: Candidate trajectories for the UAV are generated. Each candidate trajectory is scored using an improved DWA evaluation function. The linear velocity and attitude angular velocity of the candidate trajectory with the highest score are selected as the original control commands. The original control commands are smoothed to obtain the UAV control commands.

10. A method for predicting and tracking maneuvering targets of unmanned aerial vehicles based on an improved dynamic window method according to claim 9, characterized in that, The score corresponding to the improved DWA evaluation function is: Where α,β,γ,∈,ζ,η,ξ,ρ are the weight coefficients of each indicator item, and the pitch angle evaluation item is... Yaw angle evaluation item Range term for maneuvering target state estimation Predicted range term for maneuvering target status potential field virtual target point distance term Obstacle avoidance distance item High safety distance item and speed item vel k,s ; Among them, the height safety distance item for in, The high security score for the s-th candidate trajectory at time k. d represents the height at the end of the trajectory. safe Safety height threshold; Range term for maneuvering target state estimation for: in, Let S be the distance score between the s-th candidate trajectory at time k and the estimated position of the maneuvering target. Let be the distance between the s-th trajectory and the estimated position of the target. This represents the sum of the distances of all trajectories; Predicted range term for maneuvering target status for: in, Let the candidate trajectory at time k be the predicted position of the maneuvering target. Distance rating Let be the distance between the s-th trajectory and the predicted location of the target. This represents the sum of distances between all trajectories and the predicted location; potential field virtual target point distance term for: in, Let s be the candidate trajectory and the virtual target point of the potential field at time k. Distance rating Let be the distance between the s-th trajectory and the predicted location of the target. This represents the sum of the distances between all trajectories and the virtual target point in the potential field.

Images