CN115963856A - Rapid target tracking method for quad-rotor unmanned aerial vehicle - Google Patents
Rapid target tracking method for quad-rotor unmanned aerial vehicle Download PDFInfo
- Publication number
- CN115963856A CN115963856A CN202310002785.5A CN202310002785A CN115963856A CN 115963856 A CN115963856 A CN 115963856A CN 202310002785 A CN202310002785 A CN 202310002785A CN 115963856 A CN115963856 A CN 115963856A
- Authority
- CN
- China
- Prior art keywords
- target
- unmanned aerial
- aerial vehicle
- track
- quad
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 238000005096 rolling process Methods 0.000 claims abstract description 4
- 238000001514 detection method Methods 0.000 claims description 19
- 238000000605 extraction Methods 0.000 claims description 7
- 238000009499 grossing Methods 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 32
- 238000005457 optimization Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 9
- 230000001133 acceleration Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000004888 barrier function Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention provides a method for quickly tracking a target of a quad-rotor unmanned aerial vehicle, which comprises the following steps: s1, acquiring target position information of a tracked object through a neural network; s2, establishing a motion model of the tracked object, and predicting the motion track of the tracked object according to the motion model; s3, performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-algorithm to generate a locally optimal front end trail; s4, optimizing the front-end track to ensure that the track is smooth and collision-free and meet the dynamics of the unmanned aerial vehicle; and S5, repositioning the target, and rolling and updating the optimized path. The method has the advantages of good reliability for detecting the target of the unmanned aerial vehicle, good target tracking effect, high robustness and quick tracking.
Description
Technical Field
The invention relates to the technical field of unmanned aerial vehicles, in particular to a fast target tracking method for a quad-rotor unmanned aerial vehicle.
Background
The quad-rotor unmanned aerial vehicle is an unmanned aerial vehicle with small volume, high sensitivity and strong concealment, has higher thrust-weight ratio and huge acceleration capacity, and is widely applied to the fields of military reconnaissance, target tracking and the like. In the target tracking process of the unmanned aerial vehicle, due to the problems of jitter, mutual shielding between targets, visual blurring and the like in the flight of the unmanned aerial vehicle, the consistency of target tracking is influenced to a certain extent, and great difficulty is brought to tracking; in addition, when the unmanned aerial vehicle flies in low altitude, obstacles in the surrounding environment bring great threat to the safe operation of the unmanned aerial vehicle and also bring great challenge to target tracking; and because the sensors and the computing resources that the unmanned aerial vehicle can bear are limited, the speed requirements on target tracking and a track generation algorithm are extremely high. The existing target tracking scheme of the quad-rotor unmanned aerial vehicle comprises methods such as an error feedback control method, a model prediction control method, an extended Kalman filtering method and an ESDF-based trajectory optimization method.
The error feedback control method is used for tracking a target based on a visual signal, and performing feedback control by taking a tracking error in an image space as a control quantity, has high running speed and strong real-time performance, cannot identify an obstacle, has poor robustness and can only be used in an open map; the model predictive control method solves the problem of nonlinear optimization of the unmanned aerial vehicle by establishing a system model, and realizes an obstacle avoidance function, but an optimization equation of the model predictive control method is a non-convex function, and is easy to fall into a local optimal solution in the optimization process; the extended Kalman filtering method predicts the motion of a target by establishing a rough image motion model, and has poor reliability because an accurate motion model cannot be established; the ESDF-based trajectory optimization method avoids obstacles through the abundant obstacle gradient information in the ESDF, but the establishment of the ESDF map requires a large amount of calculation, so that the real-time performance of the scheme is poor.
Therefore, the existing quad-rotor unmanned aerial vehicle has poor reliability on target detection, low safety in the target tracking process and low running speed.
Disclosure of Invention
Aiming at the defects of the related technologies, the invention provides a fast target tracking method for a quad-rotor unmanned aerial vehicle, which is used for solving the problems of poor reliability of target detection, low safety of a target tracking process and low running speed of the existing quad-rotor unmanned aerial vehicle.
In order to solve the technical problem, an embodiment of the present invention provides a method for tracking a fast target of a quad-rotor unmanned aerial vehicle, where the method includes the following steps:
s1, acquiring target position information of a tracked object through a neural network;
s2, establishing a motion model of the tracked object, and predicting the motion track of the tracked object according to the motion model;
s3, performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-algorithm to generate a locally optimal front end trail;
s4, optimizing the front-end track to ensure that the track is smooth and collision-free and meet the dynamics of the unmanned aerial vehicle;
and S5, repositioning the target, and rolling and updating the optimized path.
Preferably, the step S1 specifically includes the following substeps:
performing feature extraction on a video image shot by the unmanned aerial vehicle by using a Detection transform-based target Detection algorithm;
extracting global features by using a multi-head attention mechanism of a transform to generate a fixed number of Class and Bounding box prediction graphs;
calculating the confidence degrees of all prediction images by using the FFN, and outputting the types and coordinate information of objects existing in the images;
and finally, carrying out category matching and Iout matching on the coordinate frame to obtain the coordinate of the tracked object, and adding the detection result into an observation queue.
Preferably, the step S2 specifically includes the following substeps:
describing a predicted track of a tracked object by using a uniform B spline curve;
using Q target The data in the queue was fitted to a B-spline curve.
Preferably, the step S3 specifically includes the following sub-steps:
discretizing the unmanned aerial vehicle control space;
calculating a cost function of the system;
determining a target point of Hybrid A by combining the current target observation position and the target motion prediction position;
and obtaining a local optimal path by using Hybrid A-search according to the current state of the unmanned aerial vehicle and the diagonal distance between the two points.
Preferably, the step S4 specifically includes the following sub-steps:
determining a smoothing penalty term;
determining a collision penalty term;
determining a dynamics constraint penalty;
and optimizing the cost function and solving an optimal track.
Preferably, the step S5 specifically includes the following sub-steps:
generating a path gamma from the current position of the unmanned aerial vehicle to the last observation coordinate of the target through a Hybrid A algorithm obj ；
Expanding the prediction fitting curve, and adding the prediction fitting curve into a grid map to generate gamma pre ；
Gamma-gamma is formed obj And gamma pre Connected to obtain a complete relocation path gamma r The gamma is paired by a back-end trajectory optimizer r And optimizing to obtain a time-space optimal collision-free track.
Compared with the related art, the target position information of the tracked object is obtained through the neural network; establishing a motion model of the tracked object, and predicting the motion track of the tracked object according to the motion model; performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-x algorithm to generate a locally optimal front end trail; optimizing the front-end trajectory to ensure that the trajectory is smooth and collision-free and meets the dynamics of the unmanned aerial vehicle; relocate the target and scroll update the optimized path. Make four rotor unmanned aerial vehicle target tracking position accurate like this, the orbit optimization speed is fast, has promoted the arithmetic speed of algorithm, makes unmanned aerial vehicle can roll in real time in service and generate optimal path.
Drawings
The present invention will be described in detail below with reference to the accompanying drawings. The foregoing and other aspects of the invention will become more apparent and more readily appreciated from the following detailed description, taken in conjunction with the accompanying drawings. In the drawings:
fig. 1 is a flow chart of a method for fast target tracking of a quad-rotor unmanned aerial vehicle according to the present invention;
fig. 2 is a general flowchart of the fast target tracking method of a quad-rotor unmanned aerial vehicle according to the present invention;
FIG. 3 is a diagram illustrating the DETR structure of the present invention;
FIG. 4 is a schematic diagram of target point selection according to the present invention;
FIG. 5 is a schematic diagram of collision estimation in accordance with the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
The embodiments/examples described herein are specific embodiments of the invention, are intended to be illustrative of the concepts of the invention, are exemplary and explanatory, and should not be construed as limiting the embodiments of the invention and the scope of the invention. In addition to the embodiments described herein, those skilled in the art will be able to employ other technical solutions which are obvious based on the disclosure of the claims and the specification of the present application, and these technical solutions include those which make any obvious replacement or modification of the embodiments described herein, and all of which are within the scope of the present invention.
Example one
Referring to fig. 1-2, the present invention provides a method for fast target tracking of a quad-rotor unmanned aerial vehicle, comprising the following steps:
s1, acquiring target position information of a tracking object through a neural network.
S2, establishing a motion model of the tracked object, and predicting the motion trail of the tracked object according to the motion model;
s3, performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-algorithm to generate a locally optimal front end trail;
s4, optimizing the front-end track to ensure that the track is smooth and collision-free and meet the dynamics of the unmanned aerial vehicle;
and S5, repositioning the target, and rolling and updating the optimized path.
Specifically, target position information of a tracking object is obtained through a neural network; establishing a motion model of the tracked object, and predicting the motion track of the tracked object according to the motion model; performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-x algorithm to generate a locally optimal front end trail; optimizing the front-end trajectory to ensure that the trajectory is smooth and collision-free and meets the dynamics of the unmanned aerial vehicle; relocate the target and scroll update the optimized path. Therefore, the target tracking position of the quad-rotor unmanned aerial vehicle is accurate, the track optimization speed is high, the operation speed of the algorithm is improved, and the unmanned aerial vehicle can roll in real time to generate an optimal path in operation. An accurate target motion prediction model is established through a target relocation strategy, when a tracking target of the unmanned aerial vehicle is lost, the tracking target can be relocated quickly, and the robustness of the algorithm is enhanced.
In this embodiment, as shown in fig. 3, the step S1 specifically includes the following sub-steps:
and performing feature extraction on the video image shot by the unmanned aerial vehicle by using a Detection transform-based target Detection algorithm. The target Detection method based on the Detection transform is used for target Detection, and compared with target Detection methods such as R-CNN and YOLO, the method has the advantages that end-to-end target Detection is achieved, neural network training and deployment processes are simplified, and network efficiency is improved.
And the modified ResNet50 is used as a backbone network to extract the characteristics of the input image. First, the input image is size-transformed, resulting in an input image of 3 × 576 × 1024:
x img ∈R 3*576*1024 ；
performing feature extraction on the input image through ResNet50 to generate a group of 1024 × 9 × 16 original image feature maps, reducing the number of channels of the extracted feature maps to 256 in order to reduce the number of model parameters, and obtaining a new feature map:
x fea ∈R 256*9*16 ；
then adding fixed position coding information to the new characteristic diagram:
global features are extracted by using a multi-head attention mechanism of a Transformer, and a fixed number of Class and Bounding box prediction graphs are generated.
Wherein, will
And dimension reduction is carried out, and the queue is expanded into a group of 256 × 144 to be used as the input of a transform encoder-decoder layer, so that the global feature extraction is carried out. In the Encoder process, global feature extraction is carried out on the image by using a multi-head attention mechanism, and 100 Class and Bounding box prediction graphs are output through the Decoder:
featureC,featureB∈R 100*256 。
and calculating the confidence degrees of all the prediction images by using the FFN, and outputting the object type and the coordinate information existing in the images.
Wherein, calculating the score values of all prediction images output in Step2, and removing redundant background frames. Calculating the score values of all prediction maps on 20 objects through a full connection layer:
LossC＝featureC*wightC LossB＝featureB*wightB；
where wingc and wingb are coefficients of the full connection layer. The confidence of all prediction maps for each class of objects is then calculated by the Softmax function:
wherein scoreC ij 、scoreB ij Respectively the confidence coefficient of the ith Class and Bounding box prediction images to the jth Class object, screening out the result with larger confidence coefficient through a threshold value, and outputting no packet in 100 prediction imagesThe prediction box containing the object is deleted.
And finally, carrying out category matching and Iout matching on the coordinate frame to obtain the coordinate of the tracked object, and adding the detection result into an observation queue.
Performing category matching on objects output by the Detection transducer, screening out objects of the same type as the tracked object, performing Iout matching on all objects of the same type, finding the currently tracked target, and calculating the real-time three-dimensional coordinate p of the target by combining depth information ti . Coordinate p of the object ti And a time stamp t i Adding a FIFO observation queue Q target The method comprises the following steps:
Q target ＝[q 1 ,q 2 ,…,q L ]；
wherein q is i ＝{p ti ,t i L is Q target Length of queue, t L Equal to the current time.
In this embodiment, the step S2 specifically includes the following sub-steps:
the predicted trajectory of the tracked object is described using a uniform B-spline curve.
Wherein, a uniform B-spline curve is used to describe the predicted trajectory of the tracked object, and the B-spline curve is expressed as:
wherein n is the number of control points, k represents the order of the B-spline curve, and [ p ] 0 ,p 1 ,…,p n ]Is a set of control points of the B-spline curve, B i,k (t) is a k-th order B-spline basis function, and the recursion formula is as follows:
t∈{t 1 ,t 2 ,…,t n+k }。
using Q target The data in the queue was fitted to a B-spline curve.
Wherein Q is used target The data in the queue was fitted to a B-spline curve. Generating a predicted trajectory of the target by fitting past observations when new observations of the target are obtained
Extending the trajectory to (t) L ,t L+p ]At the moment, the object motion is predicted during this period. Earlier Q over time target Confidence of the elements in the queue will decrease, so we choose the hyperbolic tangent function tanh (x) to compute the weight ω of the observed values for different timestamps ti ：
Wherein k is t Is a proportional parameter. When t is i When the time difference between the current time tL and the function value is increased, the function value is quick and small, so that the confidence degrees of different observed values can be effectively distinguished. The cost function of the predicted trajectory is expressed as follows:
wherein the first term is a position penalty term of the prediction curve, the acceleration penalty term is added to the second term for avoiding overfitting, and the optimal prediction track can be fitted through quadratic optimization
Specifically, the target is tracked more accurately by modeling the tracked target, fitting the target motion prediction track by using a B spline curve and combining the target observation position and the target motion prediction position as the track optimization target position.
In this embodiment, the step S3 specifically includes the following sub-steps:
discretizing the unmanned aerial vehicle control space.
Wherein, setting
Representing the system state of the unmanned aerial vehicle, and adding the acceleration u = [ a ] x ,a y ,a z ] T As system input. Let u e [ -a ] max ,a max ]And | a _ max | is the maximum acceleration input value. Discretizing the input u and the forward integration time Δ T:
the state transition equation of the system is:
a cost function of the system is calculated.
Wherein, in Hybrid A * In the algorithm, the cost function f = g + h. Wherein g is the sum of the cost value from the current node to the father node and the cost value from the father node to the starting point, and h is the cost value from the current node to the target point. Firstly, determining a g value, iteratively solving a total g value as long as a cost value from a current node to a parent node of the current node is solved, and carrying out node expansion under given input u and delta T, wherein the equation of g is as follows:
and determining a target point of Hybrid A by combining the current target observation position and the target motion prediction position.
Wherein, in combination withDetermining Hybrid A according to the observed position and the predicted position of the target motion * Target point x of (2) g As shown in fig. 4.
Along the edge
Expanding the target motion prediction track by p moments to obtain a target prediction state:
let the target current state be x gc Then the target point x of Hybrid A g Comprises the following steps:
wherein
For weight, the h function is defined as follows:
h＝Diag(x c ,x g )。
and obtaining a local optimal path by using Hybrid A-search according to the current state of the unmanned aerial vehicle and the diagonal distance between the two points. Wherein x c For the current state of the drone, diag () represents finding the diagonal distance of two points. And searching a local optimal path by using Hybrid A.
In this embodiment, the step S4 specifically includes the following sub-steps:
a smoothing penalty term is determined.
Wherein, the tracking track is set as an order of l, the number of control points is set as N l B spline curve of
Which is composed of N l Each control point is based on>
And a set of uniform time series t 1 ,t 2 ,…,t N Define } define->
The cost function for the optimized trajectory is:
J＝α s J s +α c J c +λ d J d ；
wherein J s 、J c 、J d Respectively representing a smooth penalty term, a collision penalty term and a dynamic constraint term, lambda s 、λ c 、λ d The weight coefficients of the terms are respectively.
A smoothing penalty term is determined. From the properties of the B-spline curve, one can obtain
Control points of the derivatives of each order, and are all { Q 1 ,Q 2 ,…,Q Nl A linear combination of.
Wherein V i 、A i 、J i Respectively as an optimized trajectory
Control points of second to fourth order derivatives. The cost function of the smooth penalty term is:
therefore, the property of the B-spline curve is effectively determined, the control points of each derivative can be conveniently obtained, the calculation is accurate, and the application range is wide.
A collision penalty term is determined.
Wherein a collision penalty term is determined. Discretizing an optimization curve intoIn the grid map, if a certain B-spline curve touches an obstacle, a new collision-free path gamma is generated by a path search algorithm. And each control point Q in the obstacle i Mapped to the surface of the obstacle by Γ, as shown in fig. 5:
b-spline curve at control point Q i The intersection point of the normal plane omega and gamma is denoted as p oi Will be along Q i Point of direction p oi The ray of (c) is denoted as l. The point of intersection of l with the nearest surface of the jth obstacle is denoted as p ij And generates a corresponding direction vector v ij Each of { p ij ,v ij That a pair belongs to only one particular control point Q i . From { p ij ,v ij The control point Q can be calculated quickly i Distance d to the surface of the obstacle ij Comprises the following steps:
d ij ＝(Q i -p ij )·v ij ；
setting the safe distance d from the track to the obstacle f To d is paired ij ＜d f The control point of (a) makes a penalty. To facilitate further optimization, a quadratic continuously differentiable metric function is constructed and is applied at d ij Suppression of its slope when decreasing:
c ij ＝d f -d ij ；
j c (i, j) is a control point Q i Is (p) ij ,v ij The cost value of the generation. All j are c (i, j) add, then the total collision penalty function:
wherein N is l Number of control points, N, for optimizing trajectory obj The number of the obstacles currently detected is obtained.
Therefore, the collision penalty function of the track equation is calculated based on the gradient collision estimation method, the calculation of the gradient information of the global map is avoided, the calculation amount is reduced, and the running speed is accelerated.
And determining a dynamics constraint penalty. Since the acceleration of the drone cannot be infinite, it is necessary to optimize the curve
The order derivatives of (a) are constrained. Based on the convex hull property of the B-spline curve, the length of the convex hull is adjusted>
And limiting the constraint points of each order of derivative to realize dynamic constraint, wherein the cost function of the dynamic constraint term is as follows:
wherein omega v 、ω a 、ω j In order to be the weight coefficient,
is a second order continuously differentiable metric function:
wherein c is r ∈{V i ,A i ,J i All coefficients are quadratic coefficients to make them satisfy the second order continuous derivation, | λ c m I is the upper limit of the kinetic constraint (lambda → 1) - )，c j For the segmentation point of the metric function, when | c j |≤|c r The value of the metric function increases sharply.
Therefore, corresponding cost functions are quantized by designing a second-order continuous differentiable piecewise measurement function which is easy to derive, the calculation amount is reduced, and the convergence speed of the cost functions is accelerated.
And optimizing the cost function and solving an optimal track.
Wherein, for the cost functionOptimizing and solving the optimal track
For J s The derivative of the B spline curve can be easily obtained due to the characteristics of the B spline curve; j. the design is a square c 、J d The terms respectively adopt second-order continuous differentiable measurement functions, so the derivative of the terms is easy to obtain. According to the property, the function is solved by using a Newton method to obtain the optimal tracking track
In this embodiment, the step S5 specifically includes the following sub-steps:
generating a path gamma from the current position of the unmanned aerial vehicle to the last observation coordinate of the target through Hybrid A algorithm obj ；
Expanding the prediction fitting curve, and adding the prediction fitting curve into a grid map to generate gamma pre ；
Gamma-gamma is formed obj And Γ pre Connected to obtain a complete relocation path t r The gamma is paired by a back-end trajectory optimizer r And optimizing to obtain a time-space optimal collision-free track.
In particular, fast relocation after target loss. Due to the shielding of obstacles, the limitation of a perception range and the uncertainty of target movement in the tracking process, the unmanned aerial vehicle is difficult to locate the target all the time. When the drone loses the target, the target should be quickly relocated.
Generating a path gamma from the current position of the unmanned aerial vehicle to the last observation coordinate of the target through Hybrid A algorithm obj 。
Expanding the prediction fitting curve generated in the step S2, and adding the expansion fitting curve into the grid map to generate gamma pre . If f is pre On gridWhen an obstacle is encountered in the process of formatting, a new collision-free path gamma is generated through a Hybrid A-algorithm pre 。
Will gamma obj And Γ pre Connected to obtain a complete relocation path gamma r By the back end trajectory optimizer pair F in the fourth step r And optimizing to obtain a time-space optimal collision-free track. The collision section track is guided through a collision-free track, soft constraint is generated to push the collision track out of the barrier, and therefore useless gradient information is avoided being calculated, and the algorithm running speed is increased. The cost function is calculated by adopting a second-order continuous differentiable sectional type measurement function, so that the derivative of the cost function can be conveniently solved, the numerical solution is conveniently carried out, and the track optimization speed is improved.
The invention has the following technical effects:
(1) By introducing the encoder-decoder of the Transformer in the characteristic extraction stage, the end-to-end unmanned aerial vehicle target detection is realized, non-maximum value inhibition or anchor link is avoided, and the training and deployment process of the target detection algorithm is simplified.
(2) Through a light-weight rapid unmanned aerial vehicle target tracking trajectory optimization algorithm, an optimization function which is convenient to derive is adopted, numerical solution can be rapidly carried out, and smooth and collision-free local optimal paths which meet the dynamics of the unmanned aerial vehicle are calculated in real time.
(3) Through a target relocation strategy, an accurate target motion prediction model is established, when the tracking target of the unmanned aerial vehicle is lost, the tracking target can be relocated quickly, and the robustness of the algorithm is enhanced.
(4) By adopting the gradient-based collision estimation algorithm, a large amount of useless ESDF map gradient information is avoided being calculated, the operation speed of the algorithm is improved, and the unmanned aerial vehicle can roll in real time to generate an optimal path in operation.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (6)
1. A method for fast target tracking of a quad-rotor unmanned aerial vehicle, the method comprising the steps of:
s1, acquiring target position information of a tracked object through a neural network;
s2, establishing a motion model of the tracked object, and predicting the motion track of the tracked object according to the motion model;
s3, performing path search by combining the current target observation position and the motion trail prediction position through a Hybrid A-algorithm to generate a locally optimal front end trail;
s4, optimizing the front-end track to ensure that the track is smooth and collision-free and meet the dynamics of the unmanned aerial vehicle;
and S5, repositioning the target, and rolling and updating the optimized path.
2. A method for fast target tracking for quad-rotor unmanned aerial vehicles according to claim 1, wherein step S1 comprises the following substeps:
performing feature extraction on a video image shot by the unmanned aerial vehicle by using a Detection transform-based target Detection algorithm;
extracting global features by using a multi-head attention mechanism of a transform to generate a fixed number of Class and Bounding box prediction graphs;
calculating the confidence degrees of all prediction images by using the FFN, and outputting the object type and coordinate information existing in the images;
and finally, carrying out category matching and Iout matching on the coordinate frame to obtain the coordinate of the tracked object, and adding the detection result into an observation queue.
3. A method for fast target tracking for quad-rotor unmanned aerial vehicles according to claim 1, wherein step S2 comprises the following substeps:
describing a predicted track of a tracked object by using a uniform B spline curve;
using Q target The data in the queue was fitted to a B-spline curve.
4. A method for fast target tracking for quad-rotor unmanned aerial vehicles according to claim 1, wherein step S3 comprises the following substeps:
discretizing the unmanned aerial vehicle control space;
calculating a cost function of the system;
determining a target point of Hybrid A by combining the current target observation position and the target motion prediction position;
and obtaining a local optimal path by using Hybrid A-search according to the current state of the unmanned aerial vehicle and the diagonal distance between the two points.
5. A method for fast target tracking of quad-rotor unmanned aerial vehicle according to claim 4, wherein the step S4 comprises the following substeps:
determining a smoothing penalty term;
determining a collision penalty term;
determining a dynamics constraint penalty;
and optimizing the cost function and solving an optimal track.
6. A method for fast target tracking of quad-rotor unmanned aerial vehicle according to claim 3, wherein step S5 comprises the following substeps:
generating a path gamma from the current position of the unmanned aerial vehicle to the last observation coordinate of the target through Hybrid A algorithm obj ；
Expanding the prediction fitting curve, and adding the prediction fitting curve into a grid map to generate gamma pre ；
Gamma-gamma is formed obj And Γ pre Connected to obtain a complete relocation path gamma r By a back-end trajectory optimizer to r And optimizing to obtain a space-time optimal collision-free track.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310002785.5A CN115963856B (en) | 2023-01-03 | 2023-01-03 | Rapid target tracking method for four-rotor unmanned aerial vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310002785.5A CN115963856B (en) | 2023-01-03 | 2023-01-03 | Rapid target tracking method for four-rotor unmanned aerial vehicle |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115963856A true CN115963856A (en) | 2023-04-14 |
| CN115963856B CN115963856B (en) | 2024-05-10 |
Family
ID=87357647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310002785.5A Active CN115963856B (en) | 2023-01-03 | 2023-01-03 | Rapid target tracking method for four-rotor unmanned aerial vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115963856B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116679755A (en) * | 2023-06-30 | 2023-09-01 | 四川大学 | Multi-agent cooperative burst anti-striking method based on distributed model predictive control |
| CN117406771A (en) * | 2023-10-17 | 2024-01-16 | 武汉大学 | Efficient autonomous exploration method, system and equipment based on four-rotor unmanned aerial vehicle |
| CN117631686A (en) * | 2023-12-07 | 2024-03-01 | 浙江大学 | Path optimization method and track tracking control method for multi-rotor unmanned aerial vehicle |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106406359A (en) * | 2016-08-30 | 2017-02-15 | 南京航空航天大学 | Virtual object-based guidance method used for fixed wing unmanned aerial vehicle to track ground object |
| CN110632941A (en) * | 2019-09-25 | 2019-12-31 | 北京理工大学 | Trajectory generation method for target tracking of unmanned aerial vehicle in complex environment |
| CN112639815A (en) * | 2020-03-27 | 2021-04-09 | 深圳市大疆创新科技有限公司 | Target tracking method, target tracking apparatus, movable platform, and storage medium |
| CN113848982A (en) * | 2021-10-28 | 2021-12-28 | 西北工业大学太仓长三角研究院 | Method for planning and tracking control of perching and stopping moving track of quad-rotor unmanned aerial vehicle |
| CN114049602A (en) * | 2021-10-29 | 2022-02-15 | 哈尔滨工业大学 | Escape target tracking method and system based on intention reasoning |
| CN114419095A (en) * | 2021-12-13 | 2022-04-29 | 深圳先进技术研究院 | Vehicle-machine cooperative target loss tracking method, device, equipment and storage medium thereof |
| CN114740882A (en) * | 2022-03-03 | 2022-07-12 | 浙江大学湖州研究院 | Trajectory generation method for ensuring visibility of elastic target tracking by unmanned aerial vehicle |
| CN114862912A (en) * | 2022-05-09 | 2022-08-05 | 中国计量大学 | Fixed wing target tracking method based on Bezier curve secondary optimization |
-
2023
- 2023-01-03 CN CN202310002785.5A patent/CN115963856B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106406359A (en) * | 2016-08-30 | 2017-02-15 | 南京航空航天大学 | Virtual object-based guidance method used for fixed wing unmanned aerial vehicle to track ground object |
| CN110632941A (en) * | 2019-09-25 | 2019-12-31 | 北京理工大学 | Trajectory generation method for target tracking of unmanned aerial vehicle in complex environment |
| CN112639815A (en) * | 2020-03-27 | 2021-04-09 | 深圳市大疆创新科技有限公司 | Target tracking method, target tracking apparatus, movable platform, and storage medium |
| CN113848982A (en) * | 2021-10-28 | 2021-12-28 | 西北工业大学太仓长三角研究院 | Method for planning and tracking control of perching and stopping moving track of quad-rotor unmanned aerial vehicle |
| CN114049602A (en) * | 2021-10-29 | 2022-02-15 | 哈尔滨工业大学 | Escape target tracking method and system based on intention reasoning |
| CN114419095A (en) * | 2021-12-13 | 2022-04-29 | 深圳先进技术研究院 | Vehicle-machine cooperative target loss tracking method, device, equipment and storage medium thereof |
| CN114740882A (en) * | 2022-03-03 | 2022-07-12 | 浙江大学湖州研究院 | Trajectory generation method for ensuring visibility of elastic target tracking by unmanned aerial vehicle |
| CN114862912A (en) * | 2022-05-09 | 2022-08-05 | 中国计量大学 | Fixed wing target tracking method based on Bezier curve secondary optimization |
Non-Patent Citations (3)
| Title |
|---|
| YANG LIU等: "Simple Online Unmanned Aerial Vehicle Tracking with Transformer", 2021 IEEE 20TH INTERNATIONAL CONFERENCE ON TRUST, SECURITY AND PRIVACY IN COMPUTING AND COMMUNICATIONS (TRUSTCOM), pages 1236 - 1239 * |
| 王怿;祝小平;周洲;张慧;: "3维动态环境下的无人机路径跟踪算法", 机器人, no. 01 * |
| 符小卫;侯建永;高晓光;刘重;: "一种双无人机协同跟踪地面移动目标方法", 计算机应用研究, no. 07 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116679755A (en) * | 2023-06-30 | 2023-09-01 | 四川大学 | Multi-agent cooperative burst anti-striking method based on distributed model predictive control |
| CN116679755B (en) * | 2023-06-30 | 2023-12-22 | 四川大学 | Multi-agent cooperative burst anti-striking method based on distributed model predictive control |
| CN117406771A (en) * | 2023-10-17 | 2024-01-16 | 武汉大学 | Efficient autonomous exploration method, system and equipment based on four-rotor unmanned aerial vehicle |
| CN117406771B (en) * | 2023-10-17 | 2024-04-30 | 武汉大学 | Efficient autonomous exploration method, system and equipment based on four-rotor unmanned aerial vehicle |
| CN117631686A (en) * | 2023-12-07 | 2024-03-01 | 浙江大学 | Path optimization method and track tracking control method for multi-rotor unmanned aerial vehicle |
| CN117631686B (en) * | 2023-12-07 | 2024-06-07 | 浙江大学 | Path optimization method and track tracking control method for multi-rotor unmanned aerial vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115963856B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115963856B (en) | Rapid target tracking method for four-rotor unmanned aerial vehicle | |
| CN108230361B (en) | Method and system for enhancing target tracking by fusing unmanned aerial vehicle detector and tracker | |
| AU2020100371A4 (en) | Hierarchical multi-object tracking method based on saliency detection | |
| CN110660082B (en) | Target tracking method based on graph convolution and trajectory convolution network learning | |
| CN111693047B (en) | Visual navigation method for micro unmanned aerial vehicle in high-dynamic scene | |
| US7929730B2 (en) | Method and system for object detection and tracking | |
| He et al. | Vision-based UAV flight control and obstacle avoidance | |
| CN112762957B (en) | Multi-sensor fusion-based environment modeling and path planning method | |
| CN110688905B (en) | Three-dimensional object detection and tracking method based on key frame | |
| CN111968229A (en) | High-precision map making method and device | |
| CN109063549B (en) | High-resolution aerial video moving target detection method based on deep neural network | |
| CN111027505B (en) | Hierarchical multi-target tracking method based on significance detection | |
| CN116182837A (en) | Positioning and mapping method based on visual laser radar inertial tight coupling | |
| CN114719848B (en) | Unmanned aerial vehicle height estimation method based on vision and inertial navigation information fusion neural network | |
| CN114964212A (en) | Multi-machine collaborative fusion positioning and mapping method oriented to unknown space exploration | |
| CN108492324A (en) | Aircraft method for tracing based on fully-connected network and Kalman filter | |
| CN111739066B (en) | Visual positioning method, system and storage medium based on Gaussian process | |
| CN114596340A (en) | Multi-target tracking method and system for monitoring video | |
| CN117311393A (en) | Unmanned aerial vehicle autonomous flight path planning method and system | |
| CN115496900A (en) | Sparse fusion-based online carbon semantic map construction method | |
| Gong et al. | Multi-target trajectory tracking in multi-frame video images of basketball sports based on deep learning | |
| CN113064422A (en) | Autonomous underwater vehicle path planning method based on double neural network reinforcement learning | |
| CN117369507A (en) | Unmanned aerial vehicle dynamic path planning method of self-adaptive particle swarm algorithm | |
| CN114972429B (en) | Target tracking method and system for cloud edge cooperative self-adaptive reasoning path planning | |
| CN116309775A (en) | Method for fusing depth completion network based on multiple bases with SLAM |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |