An adaptive swarming method for fixed-wing UAVs using monocular vision information

Abstract

This invention discloses a method for adaptively swarming fixed-wing drones using monocular vision information, relating to the field of drone cluster control technology. The method comprises the following steps: acquiring a first-person perspective image from a monocular camera; performing target recognition on the first-person perspective image to determine the relative positions of other drones relative to the drone in the drone's body coordinate system, and calculating the absolute positions of neighboring drones in the inertial coordinate system through coordinate transformation; introducing historical visual information from fixed-wing drones to calculate the speed and direction of all drones in the neighborhood; establishing a set of visually driven behavioral rules; and constructing an adjustable weighted decision model to output the final speed and direction obtained from each round of decision-making. This invention is a communication-independent method for swarming fixed-wing drones based entirely on small-field-of-view monocular vision.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) cluster control, and in particular to a method for adaptively clustering fixed-wing UAVs using monocular vision information. Background Art

[0002] As a typical task-performing unit in artificial systems, drones (UAVs) have a wide range of practical applications, such as facility inspection and remote sensing observation. However, in complex environments, UAV swarms offer even greater advantages, significantly improving the efficiency of tasks such as aerial tracking and surveillance, coordinated search, and saturation attacks. However, most research relies on wireless RF ad hoc networks (RANs), where individual decision-making relies heavily on the local exchange of state information (such as speed, position, and survival status). This control approach is susceptible to electromagnetic interference, causing the swarm's effectiveness to rapidly degrade. Communication and sensing limitations pose significant challenges to UAV perception, decision-making, and control. To eliminate this dependency, some research is exploring the use of vision to replace traditional RF communication, thereby achieving stable self-organizing swarms.

[0003] Currently, most of the research on realizing drone swarms based on vision is done by using rotary-wing drones with more flexible camera-carrying methods. Specifically, they include:

[0004] 1) Using UV markers to mark specific locations on rotary-wing drones, each drone estimates its distance and bearing by identifying the geometric relationships between the markers. Individual drones are distinguished based on the flashing frequency of their UV LEDs, enabling relatively complex swarming behavior. This approach uses the markers to estimate local state variables, simplifying the relative positioning problem. Furthermore, the unique characteristics of the markers are exploited to identify individual drones.

[0005] 2) Use convolutional neural networks to detect and locate other individuals near the quadrotor in real time, use a multi-agent state tracker to estimate the relative position and velocity of nearby drones, and input this data into the swarm algorithm for further control.

[0006] 3) Combining target detection and tracking algorithms, we use an enhanced cascade classifier to estimate the position and velocity of a markerless rotorcraft from a basically static observer’s perspective, and use these local state estimates to make clustering decisions.

[0007] Because fixed-wing drones lack the maneuverability and controllability of rotary-wing drones, research on this topic is limited. However, current research focuses on vertical takeoff and landing fixed-wing drones. This involves using visual perception equipment to capture image information from a pilot drone. Image processing algorithms are then used to determine the pilot drone's state information, such as its azimuth and relative distance, within the pod's coordinate system. This approach then estimates the pilot drone's speed and relative position, ultimately employing a hysteresis loop control method to control the drones' formation.

[0008] Currently, most research on vision-based drone swarming methods without communication relies on rotorcraft drones, which are more maneuverable and controllable. However, these methods have the following main drawbacks:

[0009] 1) The clustering effect is often highly coupled with the environment (distance, lighting, noise) and is affected by the recognition accuracy of the recognition algorithm. The reliability and robustness of the algorithm drop sharply when the distance between machines reaches ten meters.

[0010] 2) Many studies add additional markers to drones to estimate local state quantities and identify individuals. This results in clustering effects that are overly dependent on prior information and the marker locations and are unable to respond to unmarked individuals.

[0011] 3) Verification based on rotor platforms often leverages the advantages of sensor equipment to improve the visual perception capabilities of drones (providing drones with a large field of view or even omnidirectional vision) and simplify motion constraints.

[0012] Therefore, it is an urgent problem for those skilled in the art to propose an adaptive swarming method for fixed-wing UAVs using monocular vision information to solve the difficulties existing in the existing technology. Summary of the Invention

[0013] In view of this, the present invention provides a fixed-wing UAV adaptive swarming method using monocular vision information, which can solve the control problem of UAV clusters under non-communication conditions.

[0014] In order to achieve the above object, the present invention adopts the following technical solutions:

[0015] A method for adaptively swarming fixed-wing UAVs using monocular vision information comprises the following steps:

[0016] S101. Acquire a first-perspective image based on a monocular camera;

[0017] S201. Perform target recognition on the first-person perspective image in S101 to obtain the relative positions of other individuals relative to the aircraft in the aircraft coordinate system, and obtain the absolute position of the neighboring UAV in the inertial coordinate system through coordinate transformation;

[0018] S301. After introducing the historical visual information of fixed-wing drones, calculate the speed and direction of all drones in the neighborhood;

[0019] S401. Establish a set of visually-driven behavioral rules;

[0020] S501. Build an adjustable weighted decision model to output the final speed direction obtained in each round of decision-making.

[0021] The above method may optionally further include deploying a monocular camera on the fixed-wing UAV before S101.

[0022] Optionally, the above method, S201 specifically includes the following steps:

[0023] S2011. Estimating the local state information of other drones in the first-person perspective image based on a multi-target tracking algorithm, where the local state information includes the distance between drones and the horizontal and vertical angles of the drones relative to the center of the field of view;

[0024] S2012. Calculate the relative positions of other individuals relative to the aircraft in the body coordinate system based on the local state information in S2011, and then obtain the absolute position of the neighboring UAV in the inertial coordinate system through the coordinate transformation method.

[0025] In the above method, optionally, the specific content of S301 is:

[0026] The historical visual information of fixed-wing UAVs is introduced. Based on the current and historical state position information, the trajectories of different UAVs are fitted in real time using quasi-uniform B-splines. The speed and direction of all UAVs in the neighborhood are estimated by fitting the trajectories.

[0027] In the above method, optionally, the specific content of S401 is:

[0028] The local state quantity information is used as the input of the decision model to establish a vision-dominated behavior rule set, where the behavior rule set includes but is not limited to vision-guided rules, task-guided rules, obstacle avoidance rules, and collision avoidance rules.

[0029] The above method, optional, visual guidance rules:

[0030] by For the unit prediction time step, UAV All future states are distributed within a radius of On the superior arc, the reverse extension of the velocity vector in any state passes through Current location ;

[0031]

[0032] in, For this machine exist Always on the drone The absolute position estimate of For this machine According to the track fitting results The speed estimate, is the cruising speed.

[0033] Based on this, we add formation angle constraints to control the cluster to form complex formations and use a state prediction model to perform this operation:

[0034]

[0035]

[0036] in, is the Euclidean norm, represents the space vector angle, is the initial given formation angle, and Respectively indicate the current Absolute position at the moment, based on the formation angle Solved The predicted expected position at time, For the machine and drone The relative distance estimation of is the visual extraction quantity, indicating that the drone The deviation angle from the horizontal center axis of the first-person perspective image of the aircraft.

[0037] The above method, optional, obstacle avoidance rules:

[0038] The obstacle set is defined as , specifically, circles with different horizontal cross-section radii; adding detection distance and field of view constraints, obstacles The status information is defined as ;

[0039] in, is the center coordinate of the projected circle, and the radius is , leave a safe distance margin to ensure safety ;

[0040] The judgment variables of obstacle avoidance behavior are defined as follows:

[0041]

[0042]

[0043] in, for If the projection of , then calculate the minimum adjustment angle , Indicates that the equivalent height of the obstacle is consistent with that of the drone, the operator Represents a vector In three-dimensional space Axis rotation angle .

[0044] The above method, optional, task-oriented rules:

[0045] The speed control command is given by the following formula:

[0046]

[0047] in, represents the equivalent orientation vector of the task point, For the preset task angle, Indicates this machine Current real speed, for Earth projection.

[0048] The above method, optional, collision avoidance rules:

[0049] If the drone If the position spacing is estimated to be less than Km, it is considered a safety threat and only the pitch motion of the aircraft is restricted without controlling its lateral velocity, where K is a natural number.

[0050] In the above method, optionally, the adjustable weighted decision model in S501 modifies the original input item of the visual guidance rule into:

[0051]

[0052] Where, For neighbors The weight determined by the weight and distance function;

[0053] Behavior rules are also combined based on different weights. Visual guidance rules, obstacle avoidance rules, and collision avoidance rules are given according to preset weights. The final output of the speed direction in each round of decision-making is:

[0054]

[0055] in, 、 、 are the weights of visual guidance rules, obstacle avoidance rules, and collision avoidance rules, respectively. 、 、 、 They are the speed outputs corresponding to the four rules respectively. Indicates modulo, which in this formula represents the normalization of speed.

[0056] Through the above technical solutions, it can be seen that compared with the existing technology, the present invention provides a method for adaptive swarming of fixed-wing UAVs using monocular vision information: 1) Abandoning the dependence on communication in traditional swarming methods, a three-dimensional adaptive swarming method that relies only on vision is provided; combined with the fixed-wing UAV's own maneuverability, by adding formation sight angle constraints in cluster control, it controls itself to reach the desired area at a smoothly changing speed; 2) In response to the problems of visual error transmission and accumulation in multi-machine groups, a prediction and speed compensation mechanism is proposed to maintain the long-term stability of the group's orderliness; 3) Monocular vision information is fully utilized to estimate the UAV's direction and distance, and a spline fitting method is used to process current and historical state information and obtain speed and direction estimates, thereby reproducing the motion characteristics of all UAVs in the field of view; 4) A prediction mechanism is introduced to further expand the decision-making amount, which can provide coherent and reliable navigation for UAVs in a communication-free environment; the algorithm complexity is low and the cluster scale is easy to expand. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0058] Figure 1 A flow chart of a method for adaptively swarming fixed-wing UAVs using monocular vision information provided by the present invention;

[0059] Figure 2 A decision flow chart provided for the present invention;

[0060] Figure 3 Schematic diagram of front-end detection and track fitting provided by the present invention;

[0061] Figure 4 A schematic diagram of the visual guidance rules of the present invention;

[0062] Figure 5 The following are three common obstacle types and decision output diagrams of the present invention: Figure 5 .1 is a single obstacle type, Figure 5 .2 is a multi-obstacle type, Figure 5 .3 is a continuous obstacle type;

[0063] Figure 6 A schematic diagram of the obstacle avoidance rules provided by the present invention;

[0064] Figure 7 A schematic diagram of the task-oriented rules provided by the present invention;

[0065] Figure 8 This is a schematic diagram of the decision-making information flow of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] For small fixed-wing drones, due to their limited payload capacity, only a monocular camera is deployed. Furthermore, considering the balance between effective recognition distance and field of view, this invention uses a camera with a 52° field of view. Individual decisions are made entirely based on images captured from the camera's first-person perspective, without relying on the two-way data exchange of traditional communication modes.

[0068] Reference Figure 1 As shown, the present invention discloses a method for adaptively swarming fixed-wing UAVs using monocular vision information, comprising the following steps:

[0069] S101. Acquire a first-perspective image based on a monocular camera;

[0070] S201. Perform target recognition on the first-person perspective image in S101 to obtain the relative positions of other individuals relative to the aircraft in the aircraft coordinate system, and obtain the absolute position of the neighboring UAV in the inertial coordinate system through coordinate transformation;

[0071] S301. After introducing the historical visual information of fixed-wing drones, calculate the speed and direction of all drones in the neighborhood;

[0072] S401. Establish a set of visually-driven behavioral rules;

[0073] S501. Build an adjustable weighted decision model to output the final speed direction obtained in each round of decision-making.

[0074] Furthermore, before S101, it also included the deployment of monocular cameras on fixed-wing drones.

[0075] Furthermore, S201 specifically includes the following steps:

[0076] S2011. Estimating the local state information of other drones in the first-person perspective image based on a multi-target tracking algorithm, where the local state information includes the distance between drones and the horizontal and vertical angles of the drones relative to the center of the field of view;

[0077] S2012. Calculate the relative positions of other individuals relative to the aircraft in the body coordinate system based on the local state information in S2011, and then obtain the absolute position of the neighboring UAV in the inertial coordinate system through the coordinate transformation method.

[0078] See also Figure 2 Specifically, a local state estimation model for front-end detection is established. , the set of all neighborhood individuals that can be perceived in the front-end detection process is defined as . Assume that at some moment Perceive neighbors through vision , its visual state can be derived according to the recognition frame , 、 、 Respectively represent the horizontal and vertical angle estimation and relative distance estimation of the recognition frame relative to the center of the field of view. Using geometric relationships, we can calculate Relative to Relative position in the body coordinate system .in, 、 、 are the components of the relative position vector along the three coordinate axes of the body coordinate system.

[0079]

[0080]

[0081] For the general Unified to the inertial coordinate system, the present invention converts the drone into an inertial coordinate system in the back-end processing. Euler angles Convert to rotation matrix , and neighbors The relative position correction in the inertial coordinate system is :

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] Where, 、 、 They are the roll angle, yaw angle and pitch angle of the aircraft respectively. for The transformation matrix of . Thus, the neighbor The absolute position in the inertial coordinate system can be expressed as:

[0088]

[0089] At this point, the position estimates of all neighbors in the inertial coordinate system can be obtained. It should be noted that even if the drone and At the same time, I felt However, since the state quantities obtained by the front-end detection cannot be exactly the same, the final estimate will also be biased.

[0090] Furthermore, the specific content of S301 is:

[0091] The historical visual information of fixed-wing UAVs is introduced. Based on the current and historical state position information, the trajectories of different UAVs are fitted in real time using quasi-uniform B-splines. The speed and direction of all UAVs in the neighborhood are estimated by fitting the trajectories.

[0092] Specifically, the use of historical visual information of drones is introduced and incorporated into the distributed group decision control loop. Based on the current and historical state position information, the trajectories of different drones are fitted in real time using quasi-uniform B-splines. By fitting the trajectories, the speed and direction of all drones in the neighborhood can be estimated. Figure 3 As shown in Figure 2, by taking the derivative of the fitting function, we can obtain the velocity and direction estimates of all neighbors at the current moment.

[0093] Furthermore, the specific content of S401 is:

[0094] The local state quantity information is used as the input of the decision model to establish a vision-dominated behavior rule set, where the behavior rule set includes but is not limited to vision-guided rules, task-guided rules, obstacle avoidance rules, and collision avoidance rules.

[0095] Specifically, local state information is used as the input to the decision-making module, and a vision-driven behavioral rule set is designed, mainly including vision-guided rules, task-guided rules, obstacle avoidance rules, and collision avoidance rules. To maintain the robustness of the group, the present invention introduces a prediction mechanism at the rule level, using the future state model to solve the current optimal decision-making direction. This increases the algorithm's tolerance for front-end visual detection errors, allowing fixed-wing UAV clusters to effectively maintain group order over the long term and further develop complex group behaviors. The use of this method reasonably compromises the imbalance between computational complexity, target detection accuracy, and cluster performance.

[0096] Four behavioral rules are used to guide the drone's movement and achieve vision-driven self-organizing swarm behavior. Each behavior rule is described in detail below.

[0097] Furthermore, visual guidance rules:

[0098] by For the unit prediction time step, UAV All future states of Figure 3 The radius shown is On the superior arc, the reverse extension of the velocity vector in any state passes through Current location ;

[0099]

[0100] in, For this machine exist Always on the drone The absolute position estimate of For this machine According to the track fitting results The speed estimate, is the cruising speed;

[0101] Based on this, we add formation angle constraints to control the cluster to form complex formations and use a state prediction model to perform this operation:

[0102]

[0103]

[0104] in, is the Euclidean norm, represents the space vector angle, is the initial given formation angle, and Respectively indicate the current Absolute position at the moment, based on the formation angle Solved The predicted expected position at time, For the machine and drone The relative distance estimation of is the visual extraction quantity, indicating that the drone The deviation angle relative to the horizontal axis of the first-person perspective image of the aircraft. The corresponding visual state is as follows Figure 4 shown.

[0105] Furthermore, the obstacle avoidance rules:

[0106] The safety of drones is often a higher priority. The present invention considers single obstacle and multiple obstacle scenarios such as Figure 5 The following are three common obstacle types and decision output diagrams. Figure 5 .1 is a single obstacle type, Figure 5 .2 is a multi-obstacle type, Figure 5 .3 is the continuous obstacle type.

[0107] The obstacle set is defined as , specifically, circles with different horizontal cross-section radii; adding detection distance and field of view constraints, obstacles The status information is defined as ;

[0108] in, is the center coordinate of the projected circle, and the radius is , leave a safe distance margin to ensure safety ;

[0109] The judgment variables of obstacle avoidance behavior are defined as follows:

[0110]

[0111] in, for If the projection of , then calculate the minimum adjustment angle , Indicates that the equivalent height of the obstacle is consistent with that of the drone, the operator Represents a vector In three-dimensional space Axis rotation angle This rule ultimately outputs the current drone decision-making direction. Figure 6 Schematic diagram of obstacle avoidance rules.

[0112] Going further, task-oriented rules:

[0113] Due to the limited field of view of a monocular camera, after completing obstacle avoidance or other large maneuvers, fixed-wing drones at the front of a sub-swarm may deviate from their original path, making it difficult to observe the self-healing evolution of the swarm. In certain scenarios, the addition of this rule provides the possibility for all drones to be guided to the mission area.

[0114] The speed control command is given by the following formula:

[0115]

[0116] in, represents the equivalent orientation vector of the task point, For the preset task angle, Indicates this machine Current real speed, for If the angle between the current speed of the UAV and the vector relative to the mission point is greater than Then it deflects at a certain angle. Figure 7 The figure shows a schematic diagram of a mission-oriented rule, which only takes effect when the drone does not sense its neighbors.

[0117] Furthermore, the collision avoidance rules:

[0118] If the drone If the position spacing is estimated to be less than Km, it is considered a safety threat and only the pitch motion of the aircraft is restricted without controlling its lateral velocity, where K is a natural number.

[0119] Specifically, the monocular camera limits the ability to obtain visual information, and the drone cannot respond to individuals close to the sides or rear of the wing. Here, this study provides a simplified rule model for overview. If the estimated position distance is less than 15m, it is considered a safety threat. Only the pitch motion of the aircraft is restricted without controlling its lateral velocity. Then it is appropriately lowered, otherwise it is appropriately raised. The final output of this rule is .

[0120] Furthermore, the adjustable weighted decision model in S501 modifies the original input of the visual guidance rule into:

[0121]

[0122]

[0123] Where, For neighbors The weight determined by the weight and distance function;

[0124] Behavior rules are also combined based on different weights. Visual guidance rules, obstacle avoidance rules, and collision avoidance rules are given according to preset weights. The final output of the speed direction in each round of decision-making is:

[0125]

[0126] in, 、 、 are the weights of visual guidance rules, obstacle avoidance rules, and collision avoidance rules, respectively. 、 、 、 They are the speed outputs corresponding to the four rules respectively. Indicates modulo, which in this formula represents the normalization of speed.

[0127] As can be seen from the above formula, when the speed output of a single behavior rule is non-zero, it is normalized and weighted according to a preset ratio to output the final decision heading. Except for the obstacle status, which is given as a priori information, all other information is acquired through visual perception. It should be noted that this invention focuses on validating the grouping method and exploring the emergent mechanism of self-organizing behavior, and does not delve into the optimization of weights.

[0128] Specifically, after establishing a set of behavioral rules, the present invention constructs an adjustable weighted decision-making model. Within the visual guidance rules, the contribution of multiple visually perceived drones to the current decision is determined based on a function of weight and distance. Targets at medium distances have a greater impact on the current decision, while targets at very far or very close distances have a smaller influence.

[0129] The behavior rules are also combined based on different weights. Since the waypoint guidance rule is only effective when no other individuals or obstacles are perceived in the field of view, this allows the lead drone and the drone that has deviated from the mission area after obstacle avoidance to have the ability to correct the course. In other simulation tests, the waypoint guidance rule is not effective, and the visual guidance rule, obstacle avoidance rule, and collision avoidance rule are all used as The weight of is given.

[0130] See also Figure 8 The above is a schematic diagram of the decision information flow.

[0131] The swarming strategy described above provides an effective, lightweight means to prevent cluster dispersion in the spatial dimension. The front-end perception of behavioral rules enables drones to visually capture the richest possible cluster characteristics from a very limited information environment. However, as mission timelines lengthen, even small attitude changes in each decision-making round can lead to a gradual decline in swarm effectiveness over time. This change is slow and inconspicuous, manifesting as increasing distance from other individuals. For fixed-wing drones in particular, rule-based decision-making models cannot fully meet the requirements for emergent coordination capabilities in swarms. This invention addresses this issue by introducing speed control, using speed compensation to ensure cluster robustness in the temporal dimension. When the nearest neighbor is detected as too far away, the cruise airspeed is slightly increased, and when the distance closes, the speed is slightly reduced. Furthermore, speed compensation helps control swarm density to a certain extent. This is relatively rare in existing vision-based drone swarming research, and this invention fills this gap.

[0132] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0133] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for adaptively swarming fixed-wing drones using monocular vision information, characterized in that: The following steps are involved: S101. Acquire a first-perspective image based on a monocular camera; S201. Perform target recognition on the first-person perspective image in S101 to obtain the relative positions of other individuals relative to the aircraft in the aircraft coordinate system, and obtain the absolute position of the neighboring UAV in the inertial coordinate system through coordinate transformation; S301. After introducing the historical visual information of fixed-wing drones, calculate the speed and direction of all drones in the neighborhood; S401. Establish a set of visually-driven behavioral rules; S501. Build an adjustable weighted decision model to output the final speed direction obtained in each round of decision-making; The specific contents of S401 are: The local state information is used as the input of the decision model to establish a vision-driven behavioral rule set, which includes vision-oriented rules, task-oriented rules, obstacle avoidance rules, and collision avoidance rules. Visual guidance rules: The prediction time step is in units of τ, and the drone U i All future states are distributed on the optimal arc with a radius of v0τ. The reverse extension of the velocity vector in any state passes through U i Current location in, For this machine U i At time t, the UAV U j The absolute position estimate of For this machine U i According to the track fitting results, the j The speed estimate of v0 is the cruising speed; Based on this, we add formation angle constraints to control the cluster to form complex formations and use a state prediction model to perform this operation: Among them, ||·|| is the Euclidean norm, <·> represents the angle between space vectors, δ is the initial given formation angle, and They represent the absolute position of the aircraft at the current time t, the predicted expected position at time t+τ obtained by solving the formation angle δ, and r is the distance between the aircraft and the UAV U j The relative distance estimation of UAV U is j The deviation angle from the horizontal center axis of the first-person perspective image of the aircraft.

2. The method for adaptively swarming fixed-wing drones using monocular vision information according to claim 1, characterized in that: Prior to S101, it also included the deployment of monocular cameras on fixed-wing drones.

3. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: S201 specifically includes the following steps: S2011. Estimating the local state information of other drones in the first-person perspective image based on a multi-target tracking algorithm, where the local state information includes the distance between drones and the horizontal and vertical angles of the drones relative to the center of the field of view; S2012. Calculate the relative positions of other individuals relative to the aircraft in the body coordinate system based on the local state information in S2011, and then obtain the absolute position of the neighboring UAV in the inertial coordinate system through the coordinate transformation method.

4. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: Obstacle avoidance rules: The obstacle set is defined as O = {Q1,…,O m }, specifically, circles with different horizontal cross-section radii; adding detection distance and field of view angle constraints, obstacle O k The status information is defined as in, is the center coordinate of the projected circle, and the radius is To ensure safety, leave a safety distance margin d s ; The judgment variables of obstacle avoidance behavior are defined as follows: in, for If the projection of Then calculate the minimum adjustment angle Indicates that the equivalent height of the obstacle is consistent with that of the drone, the operator Represents the rotation angle of vector v around the z axis in three-dimensional space Indicates the local U i Current real speed.

5. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: Task-oriented rules: The speed control command is given by the following formula: Among them, p M represents the equivalent orientation vector of the task point, For the preset task angle, Indicates the local U i Current real speed, for Earth projection.

6. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: Collision avoidance rules: If the UAV U j If the position spacing is estimated to be less than Km, it is considered a safety threat and only the pitch motion of the aircraft is restricted without controlling its lateral velocity, where K is a natural number.

7. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: The specific contents of S301 are: The historical visual information of fixed-wing UAVs is introduced. Based on the current and historical state position information, the trajectories of different UAVs are fitted in real time using quasi-uniform B-splines. The speed and direction of all UAVs in the neighborhood are estimated by fitting the trajectories.

8. The method for adaptively swarming fixed-wing UAVs using monocular vision information according to claim 1, characterized in that: The adjustable weighted decision model in S501 modifies the original input of the visual guidance rule into: Where w k For U k The weight determined by the weight and distance function; Behavior rules are also combined based on different weights. Visual guidance rules, obstacle avoidance rules, and collision avoidance rules are given according to preset weights. The final output of the speed direction in each round of decision-making is: in, are the weights of visual guidance rules, obstacle avoidance rules, and collision avoidance rules, respectively. are the speed outputs corresponding to the four rules respectively, |·| represents the modulus, and in this formula represents the normalization of the speed.

Images