An unmanned aerial vehicle anti-collision system track tracking method based on dynamic probability adjustment IMM algorithm

Abstract

The present application relates to the field of unmanned aerial vehicle track tracking, and provides a kind of track tracking method of unmanned aerial vehicle anti-collision system based on dynamic probability adjustment IMM algorithm, comprising: input interaction processing, filtering processing are carried out to reactive tracker, smooth tracker and coordinated turning tracker, and the overturned coordinated turning tracker is calculated;Based on each tracker, mode probability is constructed, and each mode probability is dynamically adjusted based on penalty factor;The output interaction processing is completed by weighting fusion to each mode probability of dynamic adjustment. For the high maneuverability flight characteristics of unmanned aerial vehicle, the present application introduces the penalty factor of likelihood value update, dynamically adjusts mode probability according to the state residual of each model, realizes the intelligent switching and collaborative competition between models, solves the problem of slow response and model switching lag in the process of traditional algorithm, more accurately selects the prediction model most suitable for the current flight state, so as to improve the accuracy and stability of track tracking.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) trajectory tracking, and more specifically, to a UAV collision avoidance system trajectory tracking method based on the Dynamic Probability Adjustment (IMM) algorithm. Background Technology

[0002] With the widespread application of drones in logistics, inspection and monitoring, aerial surveying and mapping, the risk of collisions between drones and between drones and manned aircraft has increased significantly. Especially in low-altitude and complex airspace environments, aerial situational awareness and conflict avoidance have become critical issues.

[0003] Unmanned aerial vehicles (UAVs) typically rely on their onboard collision avoidance systems (CAVs) to avoid collisions. These systems integrate multiple sensors to acquire information about surrounding targets, predict target trajectories using trajectory tracking algorithms, assess potential collision risks based on the relative relationship between the UAV and the target, and generate optimal maneuvering decisions when a conflict is unacceptable to achieve obstacle avoidance and conflict evasion. Compared to traditional manned aircraft, UAVs are characterized by their small size, agile maneuverability, high acceleration, diverse flight strategies, and frequent trajectory changes. The trajectory tracking algorithms of traditional manned aircraft collision avoidance systems are no longer sufficient to meet the requirements of UAV platforms, necessitating improvements in trajectory tracking algorithm performance tailored to the specific characteristics of UAVs. Summary of the Invention

[0004] This invention aims to provide a trajectory tracking method for UAV collision avoidance systems based on the dynamically probabilistically adjusted IMM algorithm, in order to solve the problem that traditional trajectory tracking algorithms cannot meet the requirements of UAV platforms.

[0005] This invention provides a trajectory tracking method for UAV collision avoidance systems based on the dynamically probabilistically adjusted IMM algorithm, comprising: Input interaction processing is performed for reactive trackers, smooth trackers, and coordinated turn trackers; The reactive tracker, smooth tracker, and coordinated turn tracker after input interaction processing are filtered, and the flipped coordinated turn tracker is calculated. Based on reactive trackers, smooth trackers, coordinated turn trackers, and flipped coordinated turn trackers, pattern probabilities are constructed, and dynamic probability adjustments are made to each pattern probability based on a penalty factor. The output interaction process is completed by weighting and fusing the probabilities of each dynamically adjusted mode.

[0006] In a preferred embodiment, the filtering process includes: Standard Kalman filtering is used to predict reactive and smooth trackers, while unscented Kalman filtering is used to update reactive and smooth trackers. Based on the current target speed, acceleration, and angular velocity, and combined with the changes in the machine's position, an unscented Kalman filter is used to predict and update the coordinated turning tracker; Based on the current flight status, the position and orientation offset is calculated, and from this, the orientation of the coordinated turn tracker for flipping is calculated, thus obtaining the coordinated turn tracker for flipping.

[0007] In a preferred embodiment, the pattern probability includes: The mode probabilities P1 of reactive trackers and smooth trackers; The pattern probability P2 of the coordinated turn tracker and the coordinated turn tracker during rollover; The pattern probability P3 of the CV tracker and the coordinated turning tracker, where the CV tracker refers to the integrated reactive tracker and smooth tracker; The pattern probabilities P4 for reactive trackers, smooth trackers, and coordinated turn trackers.

[0008] In a preferred embodiment, the dynamic probability adjustment includes: The prior probabilities P1, P2, and P3 are obtained by multiplying the previous cycle's mode probabilities by the Markov transition matrix, multiplying them by the likelihood values ​​of each tracker under the current observation, and then normalizing them to obtain the updated posterior probabilities. Pattern probability 4 is calculated based on pattern probability P1 and pattern probability P3; Specifically, based on the flight status and the status of each tracker, the likelihood value of each tracker is dynamically adjusted by calculating different penalty factors, thereby dynamically adjusting the probability of each mode.

[0009] In a preferred embodiment, the likelihood value of the tracker in the mode probability P1 is adjusted as follows: Determine whether the speed of the smooth tracker reaches a preset threshold. If the speed reaches the preset threshold, do not adjust the likelihood value corresponding to the mode probability P1; if the speed does not reach the preset threshold, adjust the likelihood value of the smooth tracker twice. During the first adjustment, the penalty factor R1 is calculated based on the velocity difference between the reactive tracker and the smooth tracker, and the velocity covariance of the reactive tracker; the penalty factor R2 is calculated based on the angular velocity and covariance of the smooth tracker. During the second adjustment, penalty factor R3 is calculated based on the heading difference between the reactive tracker and the smooth tracker, and the velocity covariance of the reactive tracker; penalty factor R4 is calculated based on the angular velocity difference between the smooth tracker and the reactive tracker, and the angular velocity covariance of the reactive tracker. The likelihood values ​​of the smoothing tracker are updated based on penalty factors R1, R2, R3, and R4.

[0010] In a preferred embodiment, the likelihood value of the tracker in the mode probability P2 is adjusted as follows: The initial likelihood values ​​of the coordinated turn tracker and the flipped coordinated turn tracker are calculated using the general probability density function of the univariate normal distribution. Penalty factor R5 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker, and the heading covariance of the reactive tracker; penalty factor R6 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker when flipped, and the heading covariance of the reactive tracker; penalty factor R7 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker, and the angular velocity covariance of the reactive tracker; and penalty factor R8 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker when flipped. The initial likelihood values ​​of the coordinated turning tracker are updated based on penalty factors R5 and R7; the initial likelihood values ​​of the flipped coordinated turning tracker are updated based on penalty factors R6 and R8.

[0011] In a preferred embodiment, the likelihood value of the tracker in the mode probability P3 is adjusted as follows: The likelihood value of the CV tracker is updated based on the likelihood values ​​of the reactive tracker and the smooth tracker and the mode probability P1; Based on the conditions of whether the smoothing tracker's speed has not converged, whether the coordinated turn tracker's speed is greater than the CV tracker's, and whether it is in a straight flight state, it is determined whether the coordinated turn tracker is more stable than the CV tracker during the startup phase. If the coordinated turn tracker is relatively more stable, the likelihood value of the coordinated turn tracker is updated based on the speed difference between the CV tracker and the coordinated turn tracker, as well as the speed covariance of the coordinated turn tracker. If the CV tracker is more stable, the coordinated turn tracker is penalized: a penalty factor R9 is calculated based on the speed difference between the CV tracker and the coordinated turn tracker, and the speed covariance of the reactive tracker; a penalty factor R10 is calculated based on the angular velocity difference between the CV tracker and the coordinated turn tracker, and the angular velocity covariance of the reactive tracker; depending on whether the smoothing tracker has converged, a better tracker is selected between the reactive tracker and the smoothing tracker, and a penalty factor R11 is calculated based on the heading difference between the CV tracker and the coordinated turn tracker, and the heading covariance of the better tracker; the likelihood value of the coordinated turn tracker is updated based on penalty factors R9, R10, and R11.

[0012] In a preferred embodiment, the penalty factor is calculated using the following formula:

[0013] in, This indicates a certain attribute of a tracker or the difference in a certain attribute between two trackers; Let be the covariance of a certain tracker's corresponding attribute; This is the magnification factor.

[0014] In a preferred embodiment, the pattern probability P4 is calculated as follows:

[0015] in, This is the first element in the pattern probability P1, corresponding to the part of the reactive tracker. It is the second element in the pattern probability P1, which corresponds to the part of the smooth tracker; This is the first element in the pattern probability P3, corresponding to the portion of the CV tracker. It is the second element in the pattern probability P3, which corresponds to the part of the coordinated turning tracker.

[0016] In a preferred embodiment, the dynamically adjusted probabilities of each mode are weighted and fused to complete the output interaction processing. The calculation method is as follows:

[0017] in, The state space for weighted fusion of each tracker, This is the state space of the reactive tracker. For the state space of the smooth tracker, To coordinate the state space of the turning tracker, The state space for the coordinated turning tracker during flipping; This is the portion of the mode probability P2 corresponding to the coordinated turning tracker. This is the portion of the coordinated turn tracker corresponding to the flip in the mode probability P2; It is the first element in the pattern probability P4, i.e. , It is the second element in the pattern probability P4, i.e. , It is the 3rd element in the pattern probability P4, i.e. .

[0018] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: In response to the highly maneuverable flight characteristics of UAVs, this invention introduces a penalty factor for likelihood value updates and dynamically adjusts the mode probability based on the state residuals of each model. This enables intelligent switching and collaborative competition between models, solving the problems of slow response and delayed model switching in traditional algorithms during maneuvers. It also allows for more accurate selection of the prediction model best suited to the current flight state, thereby improving the accuracy and stability of trajectory tracking. Attached Figure Description

[0019] Figure 1 The flowchart illustrates a trajectory tracking method for a UAV collision avoidance system based on the dynamically probabilistically adjusted IMM algorithm, as provided in this embodiment of the invention.

[0020] Figure 2 This is a flowchart of dynamic probability adjustment in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] Example Current trajectory tracking algorithms include single-motion model and multi-model trajectory tracking. Single-motion model trajectory tracking, such as constant speed motion and constant turning rate motion, suffers from decreased tracking accuracy when encountering nonlinear maneuvers such as sudden turns, acceleration, and speed changes. Multi-model trajectory tracking algorithms offer better tracking stability compared to single-model algorithms and are applicable to various maneuver modes. However, traditional multi-model trajectory tracking algorithms update model probabilities solely through model prior transition matrices and measurement likelihood functions, lacking sensitivity adjustment to the discrepancies between actual observations and model biases. This results in untimely or excessively slow switching. When encountering sudden acceleration or turns by the target, traditional models often lag in updating model probabilities, failing to quickly enhance the weights of highly fit models and easily leading to increased prediction errors.

[0024] Therefore, this invention selects three tracking models for interactive fusion processing based on the flight characteristics of UAVs, and dynamically adjusts the mode probabilities of the interactive multi-model algorithm (IMM algorithm) by judging the flight state and combining residual-driven likelihood penalty, so as to make it more adaptable to the trajectory tracking requirements of UAVs with high maneuverability, thereby improving the accuracy and robustness of the collision avoidance system.

[0025] Specifically, the three tracking models include two constant speed (CV) tracking models and one constant turning rate (CT) tracking model. The constant speed tracking models are a reactive tracker and a smooth tracker, respectively, both calculated in a two-dimensional Cartesian coordinate system, with a state space of... Reactive trackers employ underdamped high-noise intensity, resulting in fast response but also high noise; smooth trackers employ overdamped low-noise intensity, resulting in slow response and smooth output. The constant turning rate tracking model is a coordinated turning tracker in polar coordinates, with the state space being... , respectively representing distance, bearing, speed, heading angle, acceleration, and angular velocity.

[0026] In view of this, such as Figure 1 As shown in the figure, an embodiment of the present invention provides a trajectory tracking method for a UAV collision avoidance system based on the dynamically probabilistically adjusted IMM algorithm, comprising: S100, Input Interaction Processing: For the three tracking models selected in this invention, input interaction processing is performed on the reactive tracker, the smooth tracker, and the coordinated turning tracker; the input interaction processing is to update the state of each tracker by weighted summation based on the probability of each mode calculated in the previous cycle.

[0027] S200, Filtering: The reactive tracker, smoothing tracker, and coordinated turning tracker after input interaction processing are filtered, and the flipped coordinated turning tracker is calculated. Specifically: (1) The standard Kalman filter is used to predict the reactive tracker and the smooth tracker, and the unscented Kalman filter is used to update the reactive tracker and the smooth tracker. (2) Based on the current target speed, acceleration and angular velocity, and combined with the change of the machine's position, the unscented Kalman filter is used to predict and update the coordinated turning tracker; (3) Calculate the position and orientation offset based on the current flight status, and then calculate the orientation of the overturned coordinated turn tracker to obtain the overturned coordinated turn tracker.

[0028] S300, Dynamic Probability Adjustment: Based on reactive trackers, smooth trackers, coordinated turn trackers, and flipped coordinated turn trackers, pattern probabilities are constructed, and dynamic probability adjustment is performed on each pattern probability based on a penalty factor.

[0029] The pattern probabilities constructed in this embodiment of the invention include: The mode probabilities P1 of reactive trackers and smooth trackers; The pattern probability P2 of the coordinated turn tracker and the coordinated turn tracker during rollover; The pattern probability P3 of the CV tracker and the coordinated turning tracker, where the CV tracker refers to the integrated reactive tracker and smooth tracker; The pattern probabilities P4 for reactive trackers, smooth trackers, and coordinated turn trackers.

[0030] like Figure 2 As shown, the dynamic probability adjustment includes: The prior probabilities P1, P2, and P3 are obtained by multiplying the previous cycle's mode probabilities by the Markov transition matrix, multiplying them by the likelihood values ​​of each tracker under the current observation, and then normalizing them to obtain the updated posterior probabilities. Pattern probability 4 is calculated based on pattern probability P1 and pattern probability P3; Specifically, based on the flight status and the status of each tracker, the likelihood value of each tracker is dynamically adjusted by calculating different penalty factors, thereby dynamically adjusting the probability of each mode.

[0031] The formula for calculating the penalty factor is as follows:

[0032] in, It indicates a certain attribute of a tracker or the difference of a certain attribute between two trackers, such as speed difference, heading difference, angular velocity difference; Let be the covariance of a certain tracker's corresponding attribute; This is the magnification factor.

[0033] The following is a detailed explanation of the dynamic adjustment of probabilities for each mode.

[0034] (1) Pattern probability P1: The likelihood value corresponding to the mode probability P1 is ,in, This represents the likelihood value of the reactive tracker. This represents the likelihood value of the smoothing tracker. The dynamic adjustment process is as follows: First, determine whether the speed of the smooth tracker has reached a preset threshold. If the speed has reached the preset threshold, then do not adjust the likelihood value corresponding to the mode probability P1. If the speed does not reach the preset threshold, the likelihood value of the smooth tracker Two adjustments were made: During the first adjustment, a penalty factor R1 is calculated based on the speed difference between the reactive and smoothed trackers and the speed covariance of the reactive tracker. This means that if the speed difference between the smoothed and reactive trackers is too large, the weight of the smoothed tracker will be reduced. A penalty factor R2 is calculated based on the angular velocity and covariance of the smoothed tracker to penalize the rate of change of the smoothed tracker's heading angle. If the change is too large, its reliability will be weakened.

[0035] In the second adjustment, penalty factor R3 is calculated based on the difference in heading angle and smoothing angle between the reactive and smoothing trackers, and the velocity covariance of the reactive tracker. Penalty factor R4 is calculated based on the difference in angular velocity between the smoothing and reactive trackers, and the angular velocity covariance of the reactive tracker. This means that if the difference in heading angle and angular velocity is too large, the smoothing weight is reduced. The rate of change of heading angle by the reactive tracker is used to determine whether the target is in a state of violent maneuvering. If the rate of change of heading angle is large, a smaller amplification factor is used to increase the penalty intensity.

[0036] Update the likelihood value of the smoothing tracker based on penalty factors R1, R2, R3, and R4. , is represented as:

[0037] in, t Indicates the moment before the update. t +1 indicates the updated time.

[0038] (2) Probability pattern P2: The likelihood value corresponding to the mode probability P2 is , This represents the likelihood value of the coordinated turning tracker in pattern probability P2. This represents the likelihood value of the coordinated turning tracker during rollover. The dynamic adjustment process is as follows: First, the initial likelihood values ​​of the coordinated turning tracker and the flipped coordinated turning tracker are calculated using the general probability density function of the univariate normal distribution:

[0039] in, To estimate the difference between the measured heading angle and the heading angle of the coordinated turn tracker or the overturned coordinated turn tracker. The covariance of the heading of the coordinated turn tracker or the coordinated turn tracker for rollover.

[0040] The initial likelihood value is then adjusted using calculated penalty factors: penalty factor R5 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker, and the heading covariance of the reactive tracker; penalty factor R6 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker when flipped, and the heading covariance of the reactive tracker; penalty factor R7 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker, and the angular velocity covariance of the reactive tracker; and penalty factor R8 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker when flipped, and the angular velocity covariance of the reactive tracker.

[0041] The initial likelihood values ​​of the coordinated turning track are updated based on penalty factors R5 and R7; the initial likelihood values ​​of the flipped coordinated turning track are updated based on penalty factors R6 and R8. The updated likelihood values ​​are... as follows:

[0042] (3) Pattern probability P3: The likelihood value corresponding to the mode probability P3 is ,in, The likelihood value of the CV tracker. This represents the likelihood value of the coordinated turning tracker in mode probability P3. The dynamic adjustment process is as follows: Based on the likelihood values ​​and mode probability P1 of the reactive tracker and the smoothed tracker, the likelihood value of the CV tracker is calculated. Update:

[0043] in, This is the part of the pattern probability P1 corresponding to the reactive tracker. This is the part of the pattern probability P1 corresponding to the smooth tracker.

[0044] Based on factors such as whether the smoothing tracker's speed has not converged, whether the coordinated turning tracker's speed is greater than the CV tracker's, and whether it is in a straight-flying state, it is determined whether the coordinated turning tracker is more stable than the CV tracker during the startup phase. If the coordinated turning tracker is relatively more stable, the likelihood value of the coordinated turning tracker is updated based on the speed difference between the CV tracker and the coordinated turning tracker, as well as the speed covariance of the coordinated turning tracker.

[0045] in, The speed difference between the CV tracker and the coordinated turning tracker. To reconcile the speed covariance of the turning tracker.

[0046] Otherwise, the CV tracker is considered more stable, and a penalty is imposed on the coordinated turning tracker: A penalty factor R9 is calculated based on the velocity difference between the CV tracker and the coordinated turning tracker, and the velocity covariance of the reactive tracker; a penalty factor R10 is calculated based on the angular velocity difference between the CV tracker and the coordinated turning tracker, and the angular velocity covariance of the reactive tracker; depending on whether the smoothing tracker converges, a better tracker is selected from the reactive and smoothing trackers, and a penalty factor R11 is calculated based on the heading difference between the CV tracker and the coordinated turning tracker, and the heading covariance of the better tracker; the likelihood value of the coordinated turning tracker is updated based on penalty factors R9, R10, and R11. The updated likelihood value of the coordinated turning tracker... Represented as:

[0047] (4) The pattern probability P4 is calculated as follows:

[0048] in, This is the first element in the pattern probability P1, corresponding to the part of the reactive tracker. It is the second element in the pattern probability P1, which corresponds to the part of the smooth tracker; This is the first element in the pattern probability P3, corresponding to the portion of the CV tracker. It is the second element in the mode probability P3, which corresponds to the part of the coordinated turn tracker; S400, Output Interaction Processing: The output interaction processing is completed by weighted fusion of the dynamically adjusted probabilities of each mode. The calculation method is as follows:

[0049] in, The state space for weighted fusion of each tracker, This is the state space of the reactive tracker. For the state space of the smooth tracker, To coordinate the state space of the turning tracker, The state space for the coordinated turning tracker during flipping; This is the portion of the mode probability P2 corresponding to the coordinated turning tracker. This is the portion of the coordinated turn tracker corresponding to the flip in the mode probability P2; It is the first element in the pattern probability P4, i.e. , It is the second element in the pattern probability P4, i.e. , It is the 3rd element in the pattern probability P4, i.e. .

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A trajectory tracking method for a UAV collision avoidance system based on the dynamically probabilistically adjusted IMM algorithm, characterized in that, include: Input interaction processing is performed for reactive trackers, smooth trackers, and coordinated turn trackers; The reactive tracker, smooth tracker, and coordinated turn tracker after input interaction processing are filtered, and the flipped coordinated turn tracker is calculated. Based on reactive trackers, smooth trackers, coordinated turn trackers, and flipped coordinated turn trackers, pattern probabilities are constructed, and dynamic probability adjustments are made to each pattern probability based on a penalty factor. The output interaction process is completed by weighting and fusing the probabilities of each dynamically adjusted mode. The pattern probabilities include: The mode probabilities P1 of reactive trackers and smooth trackers; The pattern probability P2 of the coordinated turn tracker and the coordinated turn tracker during rollover; The pattern probability P3 of the CV tracker and the coordinated turning tracker, where the CV tracker refers to the integrated reactive tracker and smooth tracker; The pattern probabilities P4 of reactive trackers, smooth trackers, and coordinated turning trackers; The dynamic probability adjustment includes: The prior probabilities P1, P2, and P3 are obtained by multiplying the previous cycle's mode probabilities by the Markov transition matrix, multiplying them by the likelihood values ​​of each tracker under the current observation, and then normalizing them to obtain the updated posterior probabilities. Pattern probability 4 is calculated based on pattern probability P1 and pattern probability P3; Specifically, based on the flight status and the status of each tracker, the likelihood value of each tracker is dynamically adjusted by calculating different penalty factors, thereby dynamically adjusting the probability of each mode.

2. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to claim 1, characterized in that, The likelihood values ​​of the tracker in the mode probability P1 are adjusted as follows: Determine whether the speed of the smooth tracker reaches a preset threshold. If the speed reaches the preset threshold, do not adjust the likelihood value corresponding to the mode probability P1; if the speed does not reach the preset threshold, adjust the likelihood value of the smooth tracker twice. During the first adjustment, the penalty factor R1 is calculated based on the velocity difference between the reactive tracker and the smooth tracker, and the velocity covariance of the reactive tracker; the penalty factor R2 is calculated based on the angular velocity and covariance of the smooth tracker. During the second adjustment, penalty factor R3 is calculated based on the heading difference between the reactive tracker and the smooth tracker, and the velocity covariance of the reactive tracker; penalty factor R4 is calculated based on the angular velocity difference between the smooth tracker and the reactive tracker, and the angular velocity covariance of the reactive tracker. The likelihood values ​​of the smoothing tracker are updated based on penalty factors R1, R2, R3, and R4.

3. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to claim 1, characterized in that, The likelihood values ​​of the tracker in the mode probability P2 are adjusted as follows: The initial likelihood values ​​of the coordinated turn tracker and the flipped coordinated turn tracker are calculated using the general probability density function of the univariate normal distribution. Penalty factor R5 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker, and the heading covariance of the reactive tracker; penalty factor R6 is calculated based on the heading difference between the coordinated turning tracker and the reactive tracker when flipped, and the heading covariance of the reactive tracker; penalty factor R7 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker, and the angular velocity covariance of the reactive tracker; and penalty factor R8 is calculated based on the angular velocity difference between the coordinated turning tracker and the reactive tracker when flipped. The initial likelihood values ​​of the coordinated turning tracker are updated based on penalty factors R5 and R7; the initial likelihood values ​​of the flipped coordinated turning tracker are updated based on penalty factors R6 and R8.

4. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to claim 1, characterized in that, The likelihood values ​​of the tracker in mode probability P3 are adjusted as follows: The likelihood value of the CV tracker is updated based on the likelihood values ​​of the reactive tracker and the smooth tracker and the mode probability P1; Based on the conditions of whether the smoothing tracker's speed has not converged, whether the coordinated turn tracker's speed is greater than the CV tracker's, and whether it is in a straight flight state, it is determined whether the coordinated turn tracker is more stable than the CV tracker during the startup phase. If the coordinated turn tracker is relatively more stable, the likelihood value of the coordinated turn tracker is updated based on the speed difference between the CV tracker and the coordinated turn tracker, as well as the speed covariance of the coordinated turn tracker. If the CV tracker is more stable, the coordinated turn tracker is penalized: a penalty factor R9 is calculated based on the speed difference between the CV tracker and the coordinated turn tracker, and the speed covariance of the reactive tracker; a penalty factor R10 is calculated based on the angular velocity difference between the CV tracker and the coordinated turn tracker, and the angular velocity covariance of the reactive tracker; depending on whether the smoothing tracker has converged, a better tracker is selected between the reactive tracker and the smoothing tracker, and a penalty factor R11 is calculated based on the heading difference between the CV tracker and the coordinated turn tracker, and the heading covariance of the better tracker; the likelihood value of the coordinated turn tracker is updated based on penalty factors R9, R10, and R11.

5. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to any one of claims 1-4, characterized in that, The formula for calculating the penalty factor is as follows: in, This indicates a certain attribute of a tracker or the difference in a certain attribute between two trackers; Let be the covariance of a certain tracker's corresponding attribute; This is the magnification factor.

6. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to any one of claims 1-4, characterized in that, The pattern probability P4 is calculated as follows: in, This is the first element in the pattern probability P1, corresponding to the part of the reactive tracker. It is the second element in the pattern probability P1, which corresponds to the part of the smooth tracker; This is the first element in the pattern probability P3, corresponding to the portion of the CV tracker. It is the second element in the pattern probability P3, which corresponds to the part of the coordinated turning tracker.

7. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to claim 6, characterized in that, The output interaction is completed by weighting and fusing the probabilities of each dynamically adjusted mode. The calculation method is as follows: in, The state space for weighted fusion of each tracker, This is the state space of the reactive tracker. For the state space of the smooth tracker, To coordinate the state space of the turning tracker, The state space for the coordinated turning tracker during flipping; This is the portion of the mode probability P2 corresponding to the coordinated turning tracker. This is the portion of the coordinated turn tracker corresponding to the flip in the mode probability P2; It is the first element in the pattern probability P4, i.e. , It is the second element in the pattern probability P4, i.e. , It is the 3rd element in the pattern probability P4, i.e. .

8. The UAV collision avoidance system trajectory tracking method based on the dynamically probabilistically adjusted IMM algorithm according to claim 1, characterized in that, The filtering process includes: Standard Kalman filtering is used to predict reactive and smooth trackers, while unscented Kalman filtering is used to update reactive and smooth trackers. Based on the current target speed, acceleration, and angular velocity, and combined with the changes in the machine's position, an unscented Kalman filter is used to predict and update the coordinated turning tracker; Based on the current flight status, the position and orientation offset is calculated, and from this, the orientation of the coordinated turn tracker for flipping is calculated, thus obtaining the coordinated turn tracker for flipping.

Images