CN114494339A - Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm - Google Patents
Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm Download PDFInfo
- Publication number
- CN114494339A CN114494339A CN202111639385.2A CN202111639385A CN114494339A CN 114494339 A CN114494339 A CN 114494339A CN 202111639385 A CN202111639385 A CN 202111639385A CN 114494339 A CN114494339 A CN 114494339A
- Authority
- CN
- China
- Prior art keywords
- target
- tracking
- damdnet
- unmanned aerial
- aerial vehicle
- 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 46
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000003062 neural network model Methods 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 230000033001 locomotion Effects 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 230000007704 transition Effects 0.000 claims description 2
- 238000012986 modification Methods 0.000 claims 2
- 230000004048 modification Effects 0.000 claims 2
- 230000008569 process Effects 0.000 description 9
- 238000005457 optimization Methods 0.000 description 8
- 230000007774 longterm Effects 0.000 description 5
- 230000006978 adaptation Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- RZVHIXYEVGDQDX-UHFFFAOYSA-N 9,10-anthraquinone Chemical compound C1=CC=C2C(=O)C3=CC=CC=C3C(=O)C2=C1 RZVHIXYEVGDQDX-UHFFFAOYSA-N 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm, which comprises the following steps: step 1, collecting video frames, and training a boundary box regression model according to the initial video frames through a preset tracking model; the tracking model is constructed based on a neural network model of DAMDNet and is pre-trained and initialized; step 2, removing the initial video frame, and using the residual video frame as the tracking model input to obtain the target boundary frame of the video frame and the target score corresponding to the target boundary frame; step 3, if the target score obtained in the step 2 is smaller than a preset value, updating the tracking model; and if the target score obtained in the step 2 is not less than the preset value, adjusting the optimal target boundary box by using a boundary box regression model and updating the tracking model. The invention effectively solves the problems that the tracking failure of the unmanned aerial vehicle tracking algorithm is easy and the real-time performance is poor.
Description
Technical Field
The invention relates to the technical field of target tracking, in particular to an unmanned aerial vehicle target tracking method based on a DAMDNet-EKF algorithm.
Background
Near target tracking is one of the research hotspots in the field of computer vision. The main task of target tracking is to track a specific target and obtain the information of its position, scale, etc. in each frame. At present, although the academic world and the industrial world have made exciting progress in the target tracking research, there still exist many problems and challenges, such as complex application environment, similar interference of background, change of illumination condition, shading and other external factors, and target posture change, apparent deformation, scale change, fast movement, motion blur and the like.
The classification network-based tracking method regards the tracking task as a foreground and background two-classification task, and classifies the candidate sample with the highest confidence degree in each frame. The MDNet method introduces classification into the tracking field through a network, and introduces a training mode under a multi-target field to learn robust general target characteristics aiming at the fact that ambiguity can be generated between a target and a background in a tracking scene under different videos, namely the target in the video can become a background object in other videos. However, since feature extraction needs to be performed on a large number of candidate samples, the classified tracking method has the disadvantage that the tracking speed is very slow, and real-time tracking cannot be achieved.
Disclosure of Invention
The invention aims to solve the technical problem of overcoming the defects of the prior art and provides the target tracking method of the unmanned aerial vehicle based on the DAMDNet-EKF algorithm.
The invention adopts the following technical scheme for solving the technical problems:
the unmanned aerial vehicle target tracking method based on the DAMDNet-EKF algorithm provided by the invention comprises the following steps:
step 1, collecting video frames, and training a boundary box regression model according to the initial video frames through a preset tracking model; the tracking model is constructed based on a neural network model of DAMDNet and is pre-trained and initialized;
step 2, removing the initial video frame, and using the residual video frame as the tracking model input to obtain the target boundary frame of the video frame and the target score corresponding to the target boundary frame;
step 3, if the target score obtained in the step 2 is smaller than a preset value, updating the tracking model; and if the target score obtained in the step 2 is not less than the preset value, adjusting the optimal target boundary box by using a boundary box regression model and updating the tracking model.
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the neural network model of the DAMDNet comprises an input layer, a convolution layer, a field adaptive assembly, a spatial pyramid pooling layer and a full connection layer which are sequentially connected.
The invention relates to a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, namely a learning target f of a field self-adaptive component*Comprises the following steps:
where f (-) is the objective function, i.e. x in the current frameiIs the probability of the target, B denotes the number of samples, xiRepresenting the i-th candidate box, y, in the current video frameiIs xiWhen x is a label ofiIs the target time yi1, l (-) is a loss function, λ is a regularization parameter, R (-) is a regularization term, i.e., a measure of the complexity of the DAMDNet neural network model,andsamples representing the source domain and the target domain, respectively, f is the true label function over the data distribution space H,is the data distribution space of the sample.
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the convolution layer and the field adaptive assembly in the neural network model of the DAMDNet are pre-trained by using an image classification field data set ILSVRC.
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the spatial pyramid pooling layer specifically comprises the following steps:
setting the size of the output feature of the domain self-adaptive component as x, wherein x is the width of the feature, the input of the spatial pyramid pooling layer is x, the size of the convolution kernel is k, the filling is p, the step length is s, and the output feature graph x of the spatial pyramid pooling layer isjThe size is as follows:
and inputting features with different sizes by virtue of two mapping relations of point mapping and edge mapping, outputting candidate frames with the same size by the spatial pyramid pooling layer, and directly inputting the candidate frames to the full connection layer.
As a further optimization scheme of the unmanned aerial vehicle target tracking method based on the DAMDNet-EKF algorithm, in the step 3, the optimal target boundary box refers to the target boundary box with the highest target score:
wherein x is*For the target bounding box with the highest target score, xiVideo frame target bounding box, f, obtained from the output of the tracking model+(xi) Is xiThe target score of (1).
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the target boundary frame obtained in the step 2 is obtained by correction, the correction is realized by adopting a deep convolution neural network method and an EKF method; the method comprises the following specific steps:
step 2.1, mapping the target boundary frame to obtain the motion state and direction of the target to be tracked in real time;
2.2, carrying out Canny algorithm contour detection according to the motion state and direction of the target to be tracked, and solving pixel coordinates of a target boundary frame;
and 2.3, fine adjustment is carried out on the coordinates of the contour pixels of the edge of the target, and the unmanned aerial vehicle can move along with the target at any angle in the same plane.
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the mapping in the step 2.1 is as follows:
h(x′)→Hx′,Hj→H,
x′=f(x,u)→Fx+u,Fj→F
wherein, H is cross entropy loss, x 'is target bounding box predicted pixel coordinates, → is mapping, f (x, u) and H (x') are nonlinear functions, and u is predicted gaussian noise; hjTo correct the target bounding box position matrix at the moment, FjTo predict the future target bounding box position matrix, F is the target bounding box position transition matrix.
As a further optimization scheme of the unmanned aerial vehicle target tracking method of the DAMDNet-EKF algorithm, the specific processes of the steps 2.1-2.3 are as follows:
according to the real-time target boundary frame position matrix to be tracked obtained in the step 2.1, Canny algorithm contour detection is carried out on the target to be tracked, and the coordinates (x) of the ith target boundary frame pixel are solvedi,yi) Then obtaining the image centroid coordinates of the target
Wherein n is the number of pixel coordinates,represents the average of the target bounding box pixel abscissas,an average value representing the ordinate of the pixel of the target bounding box;
calculating a difference of abscissas of the target bounding box pixelsDifference of ordinate of target bounding box pixelThe unmanned aerial vehicle performs the following control:
wherein a and b are fine tuning coefficients, a>0,b>0,xmax、ymaxThe maximum values of the abscissa and the ordinate of the target bounding box pixel are respectively.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
according to the invention, a target tracking model based on MDNet improvement is adopted, and a tracking task is executed through pre-training and initialization; the hysteresis processed by the tracking algorithm is well compensated through an EKF algorithm; the problem that the tracking failure of the unmanned aerial vehicle tracking algorithm is easy to occur and the real-time performance is poor is effectively solved.
Drawings
Fig. 1 is a block diagram of an unmanned aerial vehicle target tracking control system according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a DAMDNet-EKF tracking algorithm model according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
Fig. 1 is a block diagram of an unmanned aerial vehicle target tracking control system according to an embodiment of the present invention, and as shown in fig. 1, the structure includes:
the quad-rotor unmanned aerial vehicle adopts ARM as the main processor to control the speed of four motors, utilizes DJI miscalculations to process the target image that the camera obtained, DJI miscalculations output the control command that unmanned aerial vehicle flies and send to unmanned aerial vehicle flight control system.
The vehicle-mounted camera acquires a standard video signal and transmits the standard video signal to the FGPA for image decoding and analog-to-digital conversion. And the wireless transmission module is controlled by the FGPA to transmit the field of view information of the camera to the ground station. The ground station receives the image information and decompresses it for display on the display. While the user will select the target to track on the screen. The data and control parameters are transmitted to the FGPA through a suitable tracking algorithm, and the FGPA transmits the information to the ARM master module. The main control module is used for carrying out attitude calculation to control the airplane based on the template data, the control parameters, the GPS signals and attitude information of the gyroscope, the magnetometer and the accelerometer. At the same time, the FGPA will also control the camera via the attitude signal from the ARM.
The method comprises the steps that collected video frames pass through a preset tracking model to obtain a target boundary box of the video frames and a target score corresponding to the target boundary box; the tracking model is constructed based on a DAMDNet neural network model and is obtained after pre-training and initialization.
As shown in fig. 2, the damnet neural network model sequentially includes an input layer, a convolutional layer, a domain adaptation component, an adaptive spatial pyramid pooling layer, and a full connection layer, where the input of the input layer is set to be 107 × 107 images, the convolutional layer may be set to be three conv1-conv3 layers connected in sequence, the domain adaptation component may be set to be conv2_1 and conv3_1, the adaptive spatial pyramid pooling layer may be set to be spp, the full connection layer is a binary layer domain-specific layers fc6, and each training video corresponds to one full connection layer in the training process. The general characteristics of each video in the training set can be obtained through pre-training, so that a general expression of the tracking model is obtained, and a trained conv1-conv3_1 layer is obtained.
When a tracking task is executed at the beginning each time, on the basis of a pre-trained tracking model, the tracking model initializes an fc6 layer corresponding to the current tracking task through a first collected video frame. The specific initialization process is exemplified as follows:
firstly, a first frame of video frame is collected, the real position of the first frame of video frame is determined, 1000 candidate regions meeting the requirements are generated through multi-dimensional Gaussian distribution, the intersection ratio of the candidate regions IOU (intersection over Union) is 0.6, the candidate regions are input into a tracking model, the features of domain adaptation components conv2_1 and conv3_1 are obtained, and the features are used for carrying out boundary frame regression training by combining the real position. The features of the positive and negative sample regions are next extracted for the 500 positive sample and 5000 negative sample regions generated for the first video frame. And then, carrying out iterative training on the tracking model, and randomly selecting the features of 32 positive sample regions and the features of 1024 negative sample regions in each iteration to form a mini-batch. And inputting 1024 negative sample regions into the tracking model, circulating and calculating scores, and selecting 96 highest-scoring negative sample regions from the 1024 negative samples as the difficultly-exemplified negative sample regions. And introducing the data into the system, respectively calculating the score of a positive sample region and the score of a difficult negative sample region, calculating loss by forward propagation, and finally optimizing, updating parameters and the like to obtain an initialized fc6 layer.
And if the tracking of the acquired video frame is determined to fail and the target score is less than 0.5, updating the model in a short time. The short-time updating process specifically comprises the following steps: and selecting a number of video frames which are acquired recently and correspond to the short-term updating process, such as the latest 20 video frames, and extracting the positive sample regions and the negative sample regions of the prestored 20 video frames to iteratively train the tracking model.
Actually, after the pre-stored positive sample region and negative sample region of the video frame are determined to successfully track the currently acquired video frame, forward propagation is respectively performed on the positive sample region and the negative sample region according to the obtained positive sample region and the negative sample region acquired by the target bounding box, and then the characteristics of corresponding conv2_1 and conv3_1 are obtained and stored.
The updating strategy also sets a long-term updating strategy, the long-term updating process is periodically executed on the model according to a preset long-term updating period, for example, 8 frames, 10 frames and the like, and the positive sample regions and the negative sample regions of the video frames which are collected and correspond to the long-term updating process are selected in the long-term updating process for performing iterative training on the tracking model.
The model tracking unit is used for enabling the collected video frames to pass through a preset tracking model to obtain a target boundary box of the video frames and a target score corresponding to the target boundary box; the tracking model is constructed based on a DAMDNet neural network model and is obtained after pre-training and initialization; and the model updating unit is used for updating the model in a short time if the tracking of the acquired video frame is determined to fail according to a preset updating strategy. And meanwhile, the model is updated periodically for a long time.
An EKF (extended Kalman Filter) method is used for correcting the current measured value, real-time recursive filtering is carried out, and instability of the unmanned aerial vehicle caused by violent shaking of the following target is prevented; the EKF method can make predictions about the next state, compensating better for the hysteresis of the damnet algorithm. The principle is as follows:
h(x′)→Hx′,Hj→H,
x′=f(x,u)→Fx+u,Fj→F
wherein, H is cross entropy loss, x' is target bounding box predicted pixel coordinates, → is mapping, and u is predicted Gaussian noise; hjTo correct the target bounding box position matrix at the moment, FjTo predict the position matrix of the future target bounding box, f (x, u), H (x') are non-linear functions to obtain more accurate state prediction values, and the mapped measurement values Hj、FjIs composed of
Through the processing, the motion state and the direction of the target can be obtained in real time, Canny algorithm contour detection is carried out on the target to be tracked, and the coordinates (x) of the ith target boundary frame pixel are solvedi,yi) Then obtaining the image centroid coordinates of the target
Wherein n is the number of pixel coordinates,represents the average of the target bounding box pixel abscissas,an average value representing the ordinate of the pixel of the target bounding box;
calculating a difference of abscissas of the target bounding box pixelsDifference of ordinate of target bounding box pixelThe unmanned aerial vehicle performs the following control:
wherein a and b are fine tuning coefficients, a>0,b>0,xmax、ymaxRespectively taking the maximum values of the abscissa and the ordinate of the target boundary frame pixel; x is to beeAnd yeVectorization and vector addition can convert the vector into (x)e,ye) The target tracking method has the advantages that the unmanned aerial vehicle can move along with the target at any angle in the same plane, the algorithm design of the unmanned aerial vehicle for target tracking is integrally realized, and the problems that the unmanned aerial vehicle is easy to fail when the target is tracked and is insufficient in real-time are compensated by using the prediction idea of the EKF.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (9)
1. An unmanned aerial vehicle target tracking method based on a DAMDNet-EKF algorithm is characterized by comprising the following steps:
step 1, collecting video frames, and training a boundary box regression model according to the initial video frames through a preset tracking model; the tracking model is constructed based on a neural network model of DAMDNet and is pre-trained and initialized;
step 2, removing the initial video frame, and using the residual video frame as the tracking model input to obtain the target boundary frame of the video frame and the target score corresponding to the target boundary frame;
step 3, if the target score obtained in the step 2 is smaller than a preset value, updating the tracking model; and if the target score obtained in the step 2 is not less than the preset value, adjusting the optimal target boundary box by using a boundary box regression model and updating the tracking model.
2. The method of claim 1, wherein the neural network model of DAMDNet comprises an input layer, a convolutional layer, a domain adaptive component, a spatial pyramid pooling layer, and a fully connected layer, which are sequentially connected in sequence.
3. The method of claim 2, wherein the learning objective f of the domain adaptive component is a learning objective f*Comprises the following steps:
where f (-) is the objective function, i.e. x in the current frameiIs the probability of the target, B denotes the number of samples, xiRepresenting the i-th candidate box, y, in the current video frameiIs xiWhen x is a label ofiIs the target time yi1, l (-) is a loss function, λ is a regularization parameter, R (-) is a regularization term, i.e., a measure of the complexity of the damnet neural network model,andsamples representing the source domain and the target domain, respectively, f is the true label function over the data distribution space H,is the data distribution space of the sample.
4. The method of claim 2, wherein the convolution layer and the domain adaptive component in the neural network model of damnet are pre-trained using an image classification domain dataset ILSVRC.
5. The method for unmanned aerial vehicle target tracking based on DAMDNet-EKF algorithm as claimed in claim 2, wherein the spatial pyramid pooling layer is as follows:
and if the size of the output feature of the domain self-adaptive component is x, and x is the width of the feature, the input of the spatial pyramid pooling layer is x, the size of the convolution kernel is k, the filling is p, the step length is s, and the output feature diagram x of the spatial pyramid pooling layer is xjThe size is as follows:
and inputting features with different sizes by virtue of two mapping relations of point mapping and edge mapping, outputting candidate frames with the same size by the spatial pyramid pooling layer, and directly inputting the candidate frames to the full connection layer.
6. The method for tracking the target of the unmanned aerial vehicle based on the DAMDNet-EKF algorithm as claimed in claim 1, wherein in step 3, the optimal target boundary box is the target boundary box with the highest target score:
wherein x is*For the target bounding box with the highest target score, xiVideo frame target bounding box, f, obtained from the output of the tracking model+(xi) Is xiThe target score of (1).
7. The method for unmanned aerial vehicle target tracking based on DAMDNet-EKF algorithm as claimed in claim 1, wherein the target bounding box obtained in step 2 is obtained by modification, the modification is performed by using deep convolution neural network method and using EKF method; the method comprises the following specific steps:
step 2.1, mapping the target boundary frame to obtain the motion state and direction of the target to be tracked in real time;
2.2, carrying out Canny algorithm contour detection according to the motion state and direction of the target to be tracked, and solving pixel coordinates of a target boundary frame;
and 2.3, fine adjustment is carried out on the coordinates of the contour pixels of the edge of the target, and the unmanned aerial vehicle can move along with the target at any angle in the same plane.
8. The method for unmanned aerial vehicle target tracking based on DAMDNet-EKF algorithm as claimed in claim 7, wherein the mapping in step 2.1 is specifically as follows:
h(x′)→Hx′,Hj→H,
x′=f(x,u)→Fx+u,Fj→F
wherein, H is cross entropy loss, x 'is target bounding box predicted pixel coordinates, → is mapping, f (x, u) and H (x') are nonlinear functions, and u is predicted gaussian noise; hjFor the purpose of current correctionBounding box position matrix, FjTo predict the future target bounding box position matrix, F is the target bounding box position transition matrix.
9. The target tracking method of the unmanned aerial vehicle based on DAMDNet-EKF algorithm as claimed in claim 7, wherein the specific procedures of steps 2.1-2.3 are as follows:
according to the real-time target boundary frame position matrix to be tracked obtained in the step 2.1, Canny algorithm contour detection is carried out on the target to be tracked, and the coordinates (x) of the ith target boundary frame pixel are solvedi,yi) Then obtaining the image centroid coordinates of the target
Wherein n is the number of pixel coordinates,represents the average of the target bounding box pixel abscissas,an average value representing the ordinate of the pixel of the target bounding box;
calculating a difference of abscissas of the target bounding box pixelsDifference of ordinate of target bounding box pixelThe unmanned aerial vehicle performs the following control:
wherein a and b are fine tuning coefficients, a is more than 0, b is more than 0, and xmax、ymaxThe maximum values of the abscissa and the ordinate of the target bounding box pixel are respectively.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111639385.2A CN114494339B (en) | 2021-12-29 | 2021-12-29 | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111639385.2A CN114494339B (en) | 2021-12-29 | 2021-12-29 | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114494339A true CN114494339A (en) | 2022-05-13 |
CN114494339B CN114494339B (en) | 2024-07-12 |
Family
ID=81507339
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111639385.2A Active CN114494339B (en) | 2021-12-29 | 2021-12-29 | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114494339B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117036417A (en) * | 2023-09-12 | 2023-11-10 | 南京信息工程大学 | Multi-scale transducer target tracking method based on space-time template updating |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110807795A (en) * | 2019-10-31 | 2020-02-18 | 北方工业大学 | MDnet-based unmanned aerial vehicle remote sensing target tracking method and device |
CN111738111A (en) * | 2020-06-10 | 2020-10-02 | 杭州电子科技大学 | Road extraction method of high-resolution remote sensing image based on multi-branch cascade void space pyramid |
CN113658110A (en) * | 2021-07-22 | 2021-11-16 | 西南财经大学 | Medical image identification method based on dynamic field adaptive learning |
-
2021
- 2021-12-29 CN CN202111639385.2A patent/CN114494339B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110807795A (en) * | 2019-10-31 | 2020-02-18 | 北方工业大学 | MDnet-based unmanned aerial vehicle remote sensing target tracking method and device |
CN111738111A (en) * | 2020-06-10 | 2020-10-02 | 杭州电子科技大学 | Road extraction method of high-resolution remote sensing image based on multi-branch cascade void space pyramid |
CN113658110A (en) * | 2021-07-22 | 2021-11-16 | 西南财经大学 | Medical image identification method based on dynamic field adaptive learning |
Non-Patent Citations (2)
Title |
---|
HYEONSEOB NAM 等: "Learning Multi-Domain Convolutional Neural Networks for Visual Tracking", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》, 12 December 2016 (2016-12-12), pages 4293 - 4302 * |
郭显久 等: "基于VITAL的无人机目标跟踪方法", 《辽宁师范大学学报(自然科学版)》, vol. 42, no. 2, 30 June 2019 (2019-06-30), pages 187 - 196 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117036417A (en) * | 2023-09-12 | 2023-11-10 | 南京信息工程大学 | Multi-scale transducer target tracking method based on space-time template updating |
Also Published As
Publication number | Publication date |
---|---|
CN114494339B (en) | 2024-07-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110335290B (en) | Twin candidate region generation network target tracking method based on attention mechanism | |
CN111507271B (en) | Airborne photoelectric video target intelligent detection and identification method | |
WO2020192736A1 (en) | Object recognition method and device | |
CN109087329B (en) | Human body three-dimensional joint point estimation framework based on depth network and positioning method thereof | |
JP2022515591A (en) | 3D detection method, device, medium and device of target object | |
CN113065645B (en) | Twin attention network, image processing method and device | |
CN113705769A (en) | Neural network training method and device | |
US20210027207A1 (en) | Cross-modality image generation | |
CN110009060B (en) | Robustness long-term tracking method based on correlation filtering and target detection | |
CN111998862B (en) | BNN-based dense binocular SLAM method | |
CN111105439A (en) | Synchronous positioning and mapping method using residual attention mechanism network | |
CN114119659A (en) | Multi-sensor fusion target tracking method | |
CN113359843A (en) | Unmanned aerial vehicle autonomous landing method and device, electronic equipment and storage medium | |
CN116097307A (en) | Image processing method and related equipment | |
CN114494339B (en) | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm | |
CN117572885B (en) | Night tracking method, system and related device based on thermal infrared camera of unmanned aerial vehicle | |
CN113538527B (en) | Efficient lightweight optical flow estimation method, storage medium and device | |
CN107798329A (en) | Adaptive particle filter method for tracking target based on CNN | |
Schenkel et al. | Domain adaptation for semantic segmentation using convolutional neural networks | |
CN115909110A (en) | Lightweight infrared unmanned aerial vehicle target tracking method based on Simese network | |
KR20240012426A (en) | Unconstrained image stabilization | |
CN114022516A (en) | Bimodal visual tracking method based on high rank characteristics and position attention | |
Zhu et al. | YOLO-SDLUWD: YOLOv7-based small target detection network for infrared images in complex backgrounds | |
Friedel | Event-based visual-inertial odometry using smart features | |
CN117975201B (en) | Training data generation method, device, computer equipment and storage medium |
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 |