Vision-based multi-UAV cooperative ground target positioning method
The multi-UAV collaborative method enhances target positioning accuracy by integrating multi-dimensional observations and using end-to-end filtering to reduce sensor errors, addressing limitations in existing UAV positioning technologies.
Patent Information
- Application Number
- CN202210901076.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-28
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2042-07-28
AI Technical Summary
Existing UAV ground target positioning methods face limitations in accuracy and range, particularly with passive positioning using vision sensors, due to errors in drone position and attitude information.
A vision-based multi-UAV collaborative ground target positioning method that integrates multi-dimensional observations from multiple UAVs, utilizing a geometric model and end-to-end filtering model with unscented Kalman filtering to reduce errors and achieve accurate positioning.
The method provides high accuracy and wide applicability in target positioning by integrating multi-UAV observations, reducing sensor errors through collaborative estimation, and achieving continuous and accurate target positioning.
Smart Images
Figure CN115388890B_ABST
Abstract
Description
Technical Field
[0001] The present invention mainly relates to the field of UAV technology, and in particular to a vision-based multi-UAV collaborative ground target positioning method, which can be used for a UAV cluster with visual sensors to continuously position ground targets. Background Art
[0002] In recent years, drones have been widely used in civil and military fields to perform tasks such as search and rescue in specific areas, target reconnaissance, material transportation and delivery, disaster reconnaissance, etc. An important prerequisite for drones to complete these tasks is to accurately locate ground targets.
[0003] Generally speaking, drones use their own sensors to observe targets and obtain accurate geospatial status information of the targets, such as the position and speed of the targets in a certain geographic coordinate system. This process is called target positioning.
[0004] At present, there are two main methods for UAV ground target positioning in the existing technology:
[0005] The first type is the active positioning method, which uses a laser rangefinder or other device that actively transmits signals to ground targets and measures the distance between the ground target and the UAV to achieve target positioning. This method has high positioning accuracy but is limited by the effective range of the laser rangefinder.
[0006] The second type is the passive positioning method, which uses passive sensors such as vision, combined with the position and attitude information of the drone, and uses the geometric relationship between the drone and the target to solve the target position. In this method, the drone does not actively emit radiation sources, which makes it more concealed, and the cost of visual sensors is lower. The difficulty of the passive positioning method lies in reducing the impact of the error of the drone's position and attitude information on the target positioning accuracy. Summary of the invention
[0007] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a vision-based multi-UAV collaborative ground target positioning method with simple principle, wide application range and high positioning accuracy.
[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0009] A vision-based multi-UAV collaborative ground target positioning method, comprising:
[0010] Step S1: Establish a coordinate system;
[0011] Step S2: Calculation of the initial value of the target position based on the geometric model of cooperative observation of the target by multiple UAVs: using the observation information of the target by multiple UAVs, coordinate transformation is used to unify the observations of multiple UAVs into the world coordinate system, and then the geometric model of target observation is used to calculate the initial position of the target;
[0012] Step S3: Collaborative target position and state estimation based on the end-to-end filtering model.
[0013] As a further improvement of the present invention: the coordinate system includes establishing one or more of a geodetic coordinate system, a world coordinate system, an airborne north-east coordinate system, a body coordinate system, a pod coordinate system, a camera coordinate system, and an image coordinate system.
[0014] As a further improvement of the present invention: the geodetic coordinate system {O} adopts the international earth reference system WGS-84 system; the origin of the world coordinate system {W} is the starting position of the drone, X W The axis points to geographic north, Y W The axis points to the geographic east, Z W Pointing to the center of the earth; the origin of the airborne north-east coordinate system {N} is located at the center of mass of the drone, X N , Y N , Z N Axis and world coordinate system X W , Y W , Z W The coordinate system of the drone ({B}) has its origin at the center of mass of the drone, X B The Y axis points to the direction of the drone's nose. B The axis points to the right wing of the drone, Z B Axis and X B , Y B The axes form a right-hand coordinate system pointing to the bottom of the drone; the origin of the pod coordinate system {G} is located at the geometric center of the pod base, and the X G The Y axis points to the forward direction of the drone. G The Z axis points to the right of the drone. G Axis and X G , Y G The axes form a right-handed coordinate system pointing to the bottom of the drone; the origin of the camera coordinate system {C} coincides with the optical center of the camera, and Z C The X axis points to the optical axis of the camera. C The axis points to the right, Y C The axis points downward; the origin of the image coordinate system {I} is at the center of the image, and the X I The Y axis points to the right of the image. I The axis points downward in the image, and the coordinates of the target in the image coordinate system refer to the pixel position (u, v) in the image matrix.
[0015] As a further improvement of the present invention: the step S2 comprises:
[0016] Step S201: Using the target in the image coordinate system {I i} Calculate the line of sight unit vector from the camera's optical center to the target In the camera coordinate system {C i};
[0017] Step S202: Combine the gimbal pitch and yaw Euler angles Drone roll, pitch, yaw Euler angles Calculate the unit line of sight vector of drone i observing the target in the world coordinate system {W}
[0018] Step S203: Combine the actual measurement value of the drone camera position Get the line of sight LOS i Representation of;
[0019] Step S204: Using the UAV to coordinately observe the target geometry model to solve the initial value of the target position
[0020] As a further improvement of the present invention: in step S201, the camera is regarded as a pinhole camera model, and the visual axis unit vector from the camera optical center to the target is calculated in the camera coordinate system {C i}:
[0021]
[0022] where f i and d i are the camera intrinsic parameters related to the camera focal length and pixel size.
[0023] As a further improvement of the present invention: in step S202, the unit visual axis vector of the drone i observing the target is calculated as:
[0024]
[0025] in:
[0026]
[0027]
[0028] in They are the gimbal pitch and yaw Euler angles respectively; Euler angles for roll, pitch and yaw of the drone; It is obtained from the rotation relationship from the geometric center of the pod base to the UAV, which can be obtained through external parameter calibration; is the unit diagonal matrix.
[0029] As a further improvement of the present invention: in step S203, the visual axis LOS i It is expressed as:
[0030]
[0031] in is the actual measurement of the drone camera position, It is the actual line of sight vector representation that includes observation errors and sensor noise.
[0032] As a further improvement of the present invention: in the step S3, the collaborative target position and state estimation based on the end-to-end filtering model utilizes the interactive multi-model unscented Kalman filter to establish an end-to-end target position and state estimation algorithm from the "image position of target observation" end to the "position of the target in the world coordinate system" end, so as to continuously obtain accurate target positioning results.
[0033] As a further improvement of the present invention: Step S3 comprises:
[0034] Step S301: Initialize the system to its initial state Initialize the state covariance P0, process noise covariance Q, observation noise covariance R, one-step transition probability matrix t between models, and model probability matrix q;
[0035] Step S302: Set the current time as k; for each model, calculate the initial state of the model input interaction and the covariance matrix
[0036] Step S303: For each model, perform the Sigma point calculation process respectively; for model j, use the initial state and the covariance matrix Calculate Sigma point set and the covariance weight W of the Sigma point set c and the mean weight W m ;
[0037] Step S304: For each model, perform the state prediction process respectively; for model j, use the state model to perform a one-step prediction process for all Sigma points; calculate the mean of the Sigma point set And the covariance matrix
[0038] Step S305: For each model, perform the measurement update process respectively; for model j, calculate the measured Sigma point set Calculate the mean of the measurement value Sigma point set Covariance matrix of measurement update values
[0039] Step S306: For each model, perform the gain calculation process respectively; for model j, calculate the interaction covariance matrix between the system state and the measurement Calculate the Kalman gain of the system
[0040] Step S307: For each model, perform the filter update process separately; for model j, combine the measured value at time k Computing state estimates and the covariance matrix
[0041] Step S308: For each model, calculate the likelihood function of the filter corresponding to each model; for model j, calculate the likelihood function between the target motion model represented by the model at time k and the target real motion model Calculate the probability of model j at time k
[0042] Step S309: Using the state estimates and probabilities of all models, calculate the joint state estimate after multi-model interaction at time k and the covariance matrix O k ; Using state estimates You can get the position of the target in the world coordinate system and calculate the target's geographic location
[0043] Step S310: Repeat the process of step S302 to step S309 at a time interval Δt until the target positioning task is completed.
[0044] Compared with the prior art, the advantages of the present invention are:
[0045] 1. The vision-based multi-UAV collaborative ground target positioning method of the present invention has a simple principle, a wide range of applications, and high positioning accuracy. By constructing a multi-UAV collaborative ground target positioning method, it is possible to integrate multi-dimensional observations of ground targets by multiple UAVs to reduce target positioning errors, so that UAVs can achieve accurate and continuous positioning of ground targets.
[0046] 2. The vision-based multi-UAV collaborative ground target positioning method of the present invention can solve the problem of multiple UAVs accurately positioning ground targets in a collaborative manner. By calculating the target position based on the multi-UAV collaborative observation geometry model, it is possible to use the observation of multiple UAVs to obtain a relatively accurate initial estimate of the target position. By establishing an end-to-end filtering model for collaborative target position and state estimation, it is possible to use the continuous observation of the UAV to the target, effectively reduce the impact of sensor errors on the target positioning accuracy, and obtain a continuous and accurate target position estimate. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of the process of the present invention.
[0048] Figure 2 It is a schematic diagram of the result of establishing a coordinate system in a specific application example of the present invention.
[0049] Figure 3 It is a schematic diagram of the object and image positions in the camera coordinate system in a specific application example of the present invention.
[0050] Figure 4 It is a schematic diagram of the geometric model of multi-UAV collaborative target positioning in a specific application example of the present invention. DETAILED DESCRIPTION
[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0052] The present invention first defines the state vector X and measurement vector Z of the system as:
[0053] X=[x,y,z,v x ,v y ,v z ,a x ,a y ,a z ] T
[0054] Z=[u1,v1,…,u i ,v i ,…,u n ,v n ] T
[0055] Where (x, y, z) is defined as the position of the target in the three axes of the world coordinate system (v x ,v y ,v z ) and (a x ,a y ,a z) are the speed and acceleration of the target in the three axes respectively. For UAV i in a cluster of n UAVs, (u i ,v i ) is the pixel position of the target in the image coordinate system.
[0056] For UAV i, the nonlinear observation model Z k|k-1i =h ki (X k|k-1 )for:
[0057]
[0058] in The transformation matrix from the camera optical center to the geometric center of the pod base, is the transformation matrix from the geometric center of the pod base to the origin of the UAV body coordinate system, is the change matrix from the origin of the drone body coordinate system to the origin of the airborne north-east coordinate system, is the transformation matrix from the origin of the airborne north-east coordinate system to the origin of the world coordinate system.
[0059] like Figure 1 As shown, the vision-based multi-UAV collaborative ground target positioning method of the present invention includes:
[0060] Step S1: Establish a coordinate system;
[0061] See also Figure 2 In a specific application example, the coordinate system includes establishing one or more of a geodetic coordinate system, a world coordinate system, an airborne north-east coordinate system, a body coordinate system, a pod coordinate system, a camera coordinate system, and an image coordinate system; wherein:
[0062] The geodetic coordinate system {O} adopts the international terrestrial reference system WGS-84 system;
[0063] The origin of the world coordinate system {W} is the starting position of the drone, X W The axis points to geographic north, Y W The axis points to the geographic east, Z W Pointing to the center of the earth;
[0064] The origin of the airborne north-east coordinate system {N} is located at the center of mass of the drone, X N , Y N , Z N Axis and world coordinate system X W , Y W , Z W Axis parallel;
[0065] The drone's body coordinate system ({B} origin is located at the drone's center of mass, X BThe Y axis points to the direction of the drone's nose. B The axis points to the right wing of the drone, Z B Axis and X B , Y B The axes make up the right-hand coordinate system pointing towards the bottom of the drone.
[0066] The origin of the pod coordinate system {G} is located at the geometric center of the pod base, X G The Y axis points to the forward direction of the drone. G The Z axis points to the right of the drone. G Axis and X G , Y G The axes form a right-hand coordinate system pointing to the bottom of the drone;
[0067] The origin of the camera coordinate system {C} coincides with the optical center of the camera, Z C The X axis points to the optical axis of the camera. C The axis points to the right, Y C The axis points downward;
[0068] The origin of the image coordinate system {I} is at the center of the image, X I The Y axis points to the right of the image. I The axis points downward in the image, and the coordinates of the target in the image coordinate system refer to the pixel position (u, v) in the image matrix.
[0069] Step S2: Calculation of the initial value of the target position based on the geometric model of collaborative observation of the target by multiple UAVs: Using the observation information of the target by multiple UAVs, coordinate transformation is used to unify the observations of multiple UAVs into the world coordinate system, and then the geometric model of target observation is used to calculate the initial position of the target.
[0070] Step S3: Collaborative target position and state estimation based on the end-to-end filtering model.
[0071] In a specific application example, the process of step S2 includes:
[0072] Step S201: According to formula (1), using the target in the image coordinate system {I i} Calculate the line of sight unit vector from the camera's optical center to the target In the camera coordinate system {C i}, such as Figure 3 shown.
[0073] Step S202: According to formula (2), the Euler angles of the gimbal pitch and yaw are combined Drone roll, pitch, yaw Euler angles Calculate the unit line of sight vector of drone i observing the target in the world coordinate system {W}
[0074] Step S203: According to formula (5), combined with the actual measurement value of the drone camera position Get the line of sight LOS i Step S204: According to formula (8), the target initial value is solved by using the UAV cooperative observation geometry model. like Figure 4 shown.
[0075] In the above process, the line of sight of the drone to the target is calculated, and the specific contents are as follows:
[0076] For UAV i in a cluster of n UAVs, the target is in the image coordinate system {I i}
[0077] Consider the camera as a pinhole camera model and calculate the unit vector of the visual axis from the camera optical center to the target in the camera coordinate system {C i}:
[0078]
[0079] where f i and d i are the camera intrinsic parameters related to the camera focal length and pixel size.
[0080] Calculate the representation of the unit line of sight vector of drone i observing the target in the world coordinate system {W}:
[0081]
[0082] in:
[0083]
[0084]
[0085] in They are the gimbal pitch and yaw Euler angles respectively; Euler angles for roll, pitch and yaw of the drone; It is obtained from the rotation relationship from the geometric center of the pod base to the UAV, which can be obtained through external parameter calibration; is the unit diagonal matrix.
[0086] In the above process, the target position calculation based on the multi-UAV collaborative observation geometry model is as follows:
[0087] Considering that the line of sight for target observation starts from the optical center of the drone's onboard camera and ends at the target position, the line of sight rays of multiple drones will intersect at the target point. However, under realistic conditions, observation errors and sensor noise can cause the line of sight rays to not completely intersect. In this case, the line of sight LOS i It can be expressed as:
[0088]
[0089] in is the actual measurement of the drone camera position, It is the actual line of sight vector representation that includes observation errors and sensor noise.
[0090] Assume the position of the target in the world coordinate system is but To each line of sight ray LOS i The sum of the distances D can be expressed as:
[0091]
[0092] in is the target location Projection to the line of sight ray LOS i The vertical position of .
[0093]
[0094] Substitute equation (7) into equation (6) to find the distance and D pairs respectively. and The partial derivative of and When the partial derivatives are all 0, the position estimate of the target in the world coordinate system {W} is obtained
[0095]
[0096] According to the longitude, latitude and geodetic height coordinates of the origin of the world coordinate system, the position of the target in the geodetic coordinate system {O} can be obtained.
[0097] In a specific application example, the process of step S3 includes:
[0098] Step S301: Initialize the system to its initial state using the result of step 4 in (2) Initialize the state covariance P0, process noise covariance Q, observation noise covariance R, one-step transition probability matrix t between models, and model probability matrix q.
[0099] Step S302: Assume the current time is k. According to equations (11) and (12), for each model, calculate the initial state of the model input interaction and the covariance matrix
[0100] Step S303: According to equations (15) and (16), the Sigma point calculation process is performed for each model. For model j, the initial state and the covariance matrix Calculate Sigma point set and the covariance weight W of the Sigma point set c and the mean weight W m .
[0101] Step S304: For each model, perform the state prediction process separately. For model j, according to formula (17), perform a one-step prediction process for all Sigma points using the state model. According to formulas (21) and (22), calculate the mean of the Sigma point set and the covariance matrix
[0102] Step S305: For each model, perform the measurement update process. For model j, according to formula (23), calculate the measured Sigma point set According to formula (29), calculate the mean of the measurement value Sigma point set According to formula (30), the covariance matrix of the measurement update value is
[0103] Step S306: For each model, perform the gain calculation process. For model j, according to formula (31), calculate the interaction covariance matrix between the system state and the measurement According to formula (32), calculate the Kalman gain of the system
[0104] Step S307: For each model, perform the filter update process separately. For model j, according to equations (33) and (34), combined with the measured value at time k Computing state estimates and the covariance matrix
[0105] Step S308: For each model, calculate the likelihood function of the filter corresponding to each model. For model j, according to formula (35), calculate the likelihood function between the target motion model represented by the model at time k and the target real motion model According to formula (36), calculate the probability of model j at time k
[0106] Step S309: Using the state estimates and probabilities of all models, according to equations (37) and (38), calculate the joint state estimate after multi-model interaction at time k: and the covariance matrix P k Using state estimates You can get the position of the target in the world coordinate system and calculate the target's geographic location
[0107] Step S310: Repeat the process of step S302 to step S309 at a time interval Δt until the target positioning task is completed.
[0108] In the above process, collaborative target position and state estimation based on the end-to-end filtering model: using interactive multi-model unscented Kalman filtering to establish an end-to-end target position and state estimation algorithm from the "image position of target observation" to the "position of the target in the world coordinate system", and continuously obtain accurate target positioning results. The specific contents are:
[0109] Define the system's state vector X and measurement vector Z as:
[0110] X=[x,y,z,v x ,v y ,v z ,a x ,a y ,a z ] T (9)
[0111] Z=[u1,v1,…,u i ,v i ,…,u n ,v n ] T (10)
[0112] Where (x, y, z) is defined as the position of the target in the three axes of the world coordinate system {W} (v x ,v y ,v z ) and (a x ,a y ,a z ) are the speed and acceleration of the target in the three axes respectively. For UAV i in a cluster of n UAVs, (u i ,v i ) is the target in the image coordinate system {I i}Pixel position under .
[0113] Initialize the initial state of the system based on the geometric model of multi-UAV cooperative observation of targets State covariance P0, process noise covariance Q and observation noise covariance R. Set the one-step transition probability matrix t and model probability matrix q between multiple models.
[0114] Assume there are r = 3 models in total. For model j, use its optimal state at time k-1 to estimate and the covariance matrix Calculate the initial state of the model input interaction and the covariance matrix
[0115]
[0116]
[0117] in is the one-step transition probability from model l to model j, which is:
[0118]
[0119]
[0120] For all r models, the Sigma point calculation process is performed separately. For model j, the initial state and the covariance matrix Calculate Sigma point set and the covariance weight W of the Sigma point set c and the mean weight W m :
[0121]
[0122]
[0123] Where L is the dimension of the system state, in this algorithm L = 9; λ = α 2 (L+κ)-L is the scaling factor; α determines the degree of dispersion of the Sigma point and should take a small positive value (α=0.01 in this algorithm); κ defaults to 0; β is used to describe the distribution of the system state, and β=2 is the optimal value under Gaussian morning conditions.
[0124] For all r models, the state prediction process is performed separately. For model j, a one-step prediction process is performed for all Sigma points using the state model:
[0125]
[0126] Among them A jis the system state transfer matrix of model j. There are three motion models used, namely uniform motion model, uniform acceleration motion model and uniform turning model. The state transfer matrix A of these three models 1 =A CV , A 2 =A CA , A 3 =A CT as follows:
[0127]
[0128]
[0129]
[0130] Where Δt is the time interval from time k-1 to time k, ω is the target at level X W -Y W Angular velocity of the surface target turning.
[0131] Calculate the mean of the Sigma point set and the covariance matrix
[0132]
[0133]
[0134] For all r models, the measurement update process is performed separately, where for model j, the Sigma point set of the measurement is calculated.
[0135]
[0136] For UAV i, the nonlinear observation model Z k|k-1i =h ki (X k|k-1 )for:
[0137]
[0138] in:
[0139]
[0140]
[0141]
[0142]
[0143] in The translation vector from the camera optical center to the geometric center of the pod base, is the translation vector from the geometric center of the pod base to the center of the drone, is the translation vector from the center of gravity of the drone to the origin of the world coordinate system.
[0144] Calculate the mean of the measurement value Sigma point set
[0145]
[0146] The covariance matrix of the measurement update value is:
[0147]
[0148] For all r models, the gain calculation process is performed separately. For model j, the interaction covariance matrix between the system state and the measurement for:
[0149]
[0150] Kalman gain of the system It can be calculated as:
[0151]
[0152] For all r models, the filter update process is performed separately. For model j, the Kalman gain is used The state estimate can be calculated and the covariance matrix
[0153]
[0154]
[0155] in is the measured value at time k.
[0156] For all r models, the likelihood function of the filter corresponding to each model is calculated. For model j, the likelihood function between the target motion model represented by the model and the target real motion model at time k is for:
[0157]
[0158] Then the probability of model j at time k is for:
[0159]
[0160] Using the state estimates and probabilities of all r models, calculate the joint state estimate after multi-model interaction at time k and the covariance matrix P k :
[0161]
[0162]
[0163] That is, the state estimate and covariance matrix of the algorithm at time k. Using the state estimate The position of the target in the world coordinate system {W} can be obtained and target status, and then further calculate the target's geographic location
[0164] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A vision-based multi-UAV collaborative ground target positioning method, It is characterized in that include: Step S1: Establish a coordinate system; Step S2: Calculation of the initial value of the target position based on the geometric model of the coordinated observation of the target by multiple UAVs: using the observation information of the target by multiple UAVs, coordinate transformation is used to unify the observations of multiple UAVs into the world coordinate system, and then the geometric model of the target observation is used to calculate the initial position of the target; Step S3: collaborative target position and state estimation based on the end-to-end filtering model; The collaborative target position and state estimation based on the end-to-end filtering model uses the interactive multi-model unscented Kalman filter to establish an end-to-end target position and state estimation algorithm from the "image position of the target observation" to the "position of the target in the world coordinate system" to continuously obtain accurate target positioning results; The step S2 comprises: Step S201: Using the target in the image coordinate system The position below , calculate the unit vector of the line of sight from the camera optical center to the target In the camera coordinate system The following expression; Step S202: Combine the gimbal pitch and yaw Euler angles , UAV roll, pitch, yaw Euler angles , computing drones The unit visual axis vector of the target observation in the world coordinate system The following expression ; Step S203: Combine the actual measurement value of the drone camera position , get the visual axis Representation of; Step S204: Using the UAV to coordinately observe the target geometry model to solve the initial value of the target position ; The step S3 comprises: Step S301: Initialize the system to its initial state ; Initialize state covariance , process noise covariance , observation noise covariance , one-step transition probability matrix between models and the model probability matrix ; Step S302: Set the current time to ; For each model, calculate the initial state of the model input interaction and the covariance matrix ; Step S303: For each model, perform the Sigma point calculation process respectively; , using the initial state and the covariance matrix , calculate the Sigma point set and the covariance weight of the Sigma point set and mean weight ; Step S304: For each model, execute the state prediction process respectively; for model , use the state model to perform a one-step prediction process for all Sigma points; calculate the mean of the Sigma point set and the covariance matrix ; Step S305: For each model, perform the measurement update process respectively; , calculate the measured Sigma point set ; Calculate the mean of the measurement value Sigma point set , the covariance matrix of the measurement update value ; Step S306: For each model, perform the gain calculation process respectively; , calculate the interaction covariance matrix between system state and measurement ; Calculate the Kalman gain of the system ; Step S307: For each model, perform the filtering update process respectively; , combined with Measurement value at time , calculate the state estimate and the covariance matrix ; Step S308: For each model, calculate the likelihood function of the filter corresponding to each model; ,calculate The likelihood function between the target motion model represented by the model at the moment and the target's true motion model ; Computational model exist Probability of time ; Step S309: Using the state estimates and probabilities of all models, calculate Joint state estimation after multi-model interaction at time and the covariance matrix ; Using state estimates You can get the position of the target in the world coordinate system , and calculate the target's geographic location ; Step S310: at a time interval Repeat the process from step S302 to step S309 until the target positioning task is completed.
2. The method for multi-UAV coordinated ground target positioning based on vision according to claim 1, It is characterized in that The coordinate system includes establishing one or more of a geodetic coordinate system, a world coordinate system, an airborne north-east coordinate system, a body coordinate system, a pod coordinate system, a camera coordinate system, and an image coordinate system.
3. The method for multi-UAV coordinated ground target positioning based on vision according to claim 2, It is characterized in that The geodetic coordinate system The international terrestrial reference system WGS-84 is used; the world coordinate system The origin is the starting position of the drone. The axis points to geographic north. The axis points to the geographic east. Pointing to the center of the earth; the airborne north-east coordinate system The origin of is at the center of mass of the drone. , , Axis and world coordinate system , , Axis parallel; the body coordinate system of the UAV The origin is at the center of mass of the drone, The axis points to the direction of the drone's nose. The axis points to the right wing of the drone, Axis and , The axes form a right-hand coordinate system pointing to the bottom of the drone; the pod coordinate system The origin is at the geometric center of the pod base, The axis points forward in the direction of the drone, The axis points to the right of the drone. Axis and , The axes form a right-handed coordinate system pointing to the bottom of the drone; the camera coordinate system The origin coincides with the optical center of the camera, The axis points to the direction of the camera's optical axis. The axis points to the right. The axis points downward; the image coordinate system The origin is at the center of the image, The axis points to the right of the image. The axis points downward in the image, and the coordinates of the target in the image coordinate system refer to the pixel position in the image matrix. .
4. The method for multi-UAV coordinated ground target positioning based on vision according to claim 1, It is characterized in that In step S201, the camera is regarded as a pinhole camera model, and the visual axis unit vector from the camera optical center to the target is calculated in the camera coordinate system. The following representation: in and are the camera intrinsic parameters related to the camera focal length and pixel size.
5. The method for multi-UAV coordinated ground target positioning based on vision according to claim 1, It is characterized in that In step S202, the drone is calculated The unit visual axis vector of the target observation in the world coordinate system Representation: in: in They are the gimbal pitch and yaw Euler angles respectively; Euler angles for roll, pitch and yaw of the drone; It is obtained from the rotation relationship from the geometric center of the pod base to the UAV, which can be obtained through external parameter calibration; is the unit diagonal matrix.
6. The method for multi-UAV coordinated ground target positioning based on vision according to claim 1, It is characterized in that In step S203, the visual axis It is expressed as: （5） in is the actual measurement of the drone camera position, It is the actual line of sight vector representation that includes observation errors and sensor noise.