CN109740742A - A kind of method for tracking target based on LSTM neural network - Google Patents
A kind of method for tracking target based on LSTM neural network Download PDFInfo
- Publication number
- CN109740742A CN109740742A CN201910041887.1A CN201910041887A CN109740742A CN 109740742 A CN109740742 A CN 109740742A CN 201910041887 A CN201910041887 A CN 201910041887A CN 109740742 A CN109740742 A CN 109740742A
- Authority
- CN
- China
- Prior art keywords
- data
- target
- neural network
- information
- parameter
- 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.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 66
- 238000000034 method Methods 0.000 title claims abstract description 46
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 10
- 230000015654 memory Effects 0.000 claims abstract description 3
- 238000012360 testing method Methods 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 238000013461 design Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000002224 dissection Methods 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 238000012886 linear function Methods 0.000 claims description 3
- 210000004218 nerve net Anatomy 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 16
- 238000001914 filtration Methods 0.000 abstract description 10
- 238000004364 calculation method Methods 0.000 abstract description 9
- 210000002569 neuron Anatomy 0.000 description 8
- 230000004927 fusion Effects 0.000 description 6
- 230000000306 recurrent effect Effects 0.000 description 6
- 230000001537 neural effect Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241000288105 Grus Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 150000001875 compounds Chemical group 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
A kind of method for tracking target based on LSTM neural network, belongs to target following technical field.The present invention, to complicated, nonlinear motion target following, it is low with tracking accuracy to solve the problems, such as that target following difficulty, object module are difficult to set up using long memory models (LSTM) in short-term；The data of acquisition are carried out data processing by the latitude and longitude information and velocity information for acquiring target first；Then it is designed for the LSTM neural network structure of monotrack；Adjust LSTM neural network parameter finally to realize target following.The present invention effectively simplifies nonlinear filtering process and can effectively be tracked to complicated Nonlinear Parameter；It does not need to establish target movement model and utilizes traditional filtering algorithm；The target state of subsequent time is estimated using the target motion information of history；The inner parameter of neural network is adjusted using back-propagation algorithm；The method of learning rate decaying reduces calculation amount and improves precision.
Description
Technical field
The invention belongs to target following technical fields, and in particular to a kind of target following side based on LSTM neural network
Method.
Background technique
Maneuvering target tracking is research topic more active at present, and how quickly, accurately, reliably tracking target is mesh
Mark the main purpose of tracking system design.Target following, i.e., to the information such as the position of target, speed, attitude angle carry out in real time with
Track and prediction, target following can be divided into monotrack and multiple target tracking according to target number.Most of monotrack
It is to be filtered estimation according to bayesian principle, the target following effect of complicated movement is paid no attention in the target following of this form
Think, motion model is difficult to choose and computationally intensive.And in practical applications the movement of target be it is complicated and changeable, establish a reason
The motion model thought is very difficult and nonlinear filtering calculation amount is very big.Therefore, one is sought simply, accurately, reliably
Method for tracking target is that have great theory and practice value.Currently, by Application of Neural Network in the text of target tracking domain
Offer few, the representative fusion of such as document [1] Wei Shouhui Multisensor Target Information and the south tracking research [D]
Capital: Nanjing Aero-Space University's master thesis, BP neural network and RBF neural are applied to information fusion by 2005.
In algorithm and Track Fusion, by simulating, verifying BP neural network and RBF neural validity, and by score
Analysis discovery RBF neural has the advantages that the training time is few and noise resisting ability is strong.Document [2] Xiao Linxia is based on convolution mind
Motion target tracking through network studies the Qingdao [D]: University Of Science and Technology Of Shandong's master thesis, 2017. analyze EKF first
The performance that filtering algorithm, UKF filtering algorithm and CKF filtering algorithm track Nonlinear Parameter is proposed based on WAVELET FUZZY mind
Interacting multiple model algorithm through network simultaneously verifies its feasibility.A kind of Intelligent target tracking of the such as document [3] Cui Yaqi, Xiong Wei
Method universal design frame [P] China: 20181004494.4.2018-01-16., which is proposed, to be corresponded to Trajectory Prediction manually
Time series forecasting problem in smart field is instructed using the Recognition with Recurrent Neural Network in deep learning field using Trajectory Prediction
Practice data set, training generates preliminary Trajectory Prediction method；A boat association is corresponded to the judgement of the classification in artificial intelligence field
Problem；Track filtering is corresponded to the model parameter problem concerning study in artificial intelligence field.The document is to propose to utilize people
The universal design frame of the multiple target tracking of work intelligent method.The such as document [4] Cui Yaqi, Xiong Wei are based on Recognition with Recurrent Neural Network
Targetpath prediction technique [P] China: what 201810040500.6.2018-01-06. was mainly proposed is based on cooperation track instruction
Practice collection and radar track training set, optimization is trained to targetpath prediction loop neural network, generation matches with radar
Targetpath prediction technique.Mainly solve that existing Trajectory Prediction method model is simple, complexity is low, permeability is poor and nothing
The problem of calligraphy learning.
Existing technological deficiency: the movement of target is that have the spies such as uncertainty, erratic behavior, non-linear under actual conditions
Point, to such target following, there are motion models to be difficult to set up the problems such as low with nonlinear filtering precision.Existing neural network
It is commonly applied in the fields such as domain of data fusion, data correlation field and the computer visual image processing of target following.In recent years
In terms of gradually expanding to targetpath prediction, this belongs to radar target tracking field.Such as document [1] by BP neural network and
RBF neural is applied in information fusion algorithm and Track Fusion；Document [2] proposes the movement of convolutional neural networks
Target tracking algorism, this method are applied in computer vision nerve field；It proposes in document [3] and is existed using artificial intelligence
The universal design frame of target tracking domain, this document propose will recycle neural Application of Neural Network in the side of Trajectory Prediction
Method, but it is unspecified how to neural network progress arameter optimization, if use gradient descent algorithm or back-propagation algorithm
Arameter optimization is carried out, to reduce calculation amount and avoid that over-fitting occurs.What document [4] mainly proposed is based on cooperation track
Training set and radar track training set are trained optimization to targetpath prediction loop neural network, generate and radar phase
The targetpath prediction technique matched.The simple cycle in Recognition with Recurrent Neural Network structure relevant to time series is used in document
Neural network (SRNs), thresholding Recognition with Recurrent Neural Network (GRUs), long short-term memory Recognition with Recurrent Neural Network (LSTM) and enhancing circulation
Neural network (augmented RNN), method are to select optimum structure from above-mentioned four kinds of Recognition with Recurrent Neural Network structures to match
Corresponding complexity track target prediction.The same document does not adjust ginseng using back-propagation algorithm to above-mentioned neural network structure
Number is kept loss function small as far as possible and is not reduced calculation amount using learning rate damped system and improve precision.
Summary of the invention
The purpose of the present invention is to provide a kind of method for tracking target based on LSTM neural network, utilize long short-term memory
Model (LSTM) solves target following difficulty, object module is difficult to set up and tracks to complicated, nonlinear motion target following
The low problem of precision.
The object of the present invention is achieved like this:
A kind of method for tracking target based on LSTM neural network, includes the following steps:
Step 1: acquiring the latitude and longitude information and velocity information of target, the data of acquisition are subjected to data processing；
Step 2: the LSTM neural network structure designed for monotrack；
Step 3: adjusting LSTM neural network parameter to realize target following.
First manually method of discrimination picks out error number after acquiring the latitude and longitude information and velocity information of target in the step 1
According to then pair warp and weft degree information amplifies processing first, then the normalized and data that need pair warp and weft degree to carry out data
Dissection process, then will treated that data are divided into training set and test set, the weight and partially of network structure is trained with training set
Parameter is set, test set is used to detect target tracking accuracy, and calculates target following square mean error amount.
The LSTM neural network structure designed in the step 2 is the nerve with one layer of hidden layer and multiple-input and multiple-output
Network structure；The dimension of input layer is 3, wherein being the longitude information, dimensional information and velocity information of target respectively；Hidden layer
LSTM neural network is to forget door by three δ to remove to control discarding or increase information to realize the function of forgetting or memory
Outside forgetting door, input gate and out gate, there are also input unit and output unit, door control unit is activated using sigmoid
Function, input-output unit is using tanh activation primitive；The number of concealed nodes is set as 10；The dimension of output layer is
3, activation primitive is linear function, exports longitude, dimension and the velocity information of target respectively.
LSTM Parameters of Neural Network Structure and test are adjusted using training set data in the step 3 and data detection tracks
Effect includes the following steps,
Step 3.1: chronologically feed-in training set data, 3 dimensions that every group of data are made of the longitude and latitude and speed of target
Amount；
Step 3.2: initial parameter inside setting neural network；
Step 3.3: using the method for tracking target of LSTM neural network design by the 1st of training set to the 10th data
Information is inputted as history, calculates the 11st data；It is again that the 2nd data of training set are defeated as history to the 11st data
Enter information, calculates the 12nd data, parameter later and so on, the training set parameter predicted, relatively truer training
The parameter of collection, to calculate the loss function in training set；
Step 3.4: judging whether the hundreds of the number of iterations occur carry, if executing step 3.5, otherwise skip step
3.5, execute step 3.6；
Step 3.5: parameter is adjusted as new learning rate multiplied by after rate of decay in learning rate；
Step 3.6: using the weight and offset parameter of Reverse optimization method optimization neural network；
Step 3.7: judging whether to reach the number of iterations, if so then execute step 3.8, if not then jumping to step 3.3；
Step 3.8: chronologically feed-in test set data are inputted the 1st of test set to the 10th data as history
The 2nd of test set to the 11st data are inputted the 12nd data of information prediction by the 11st data of information prediction,
Prediction data later and so on, obtaining output dimension is three-dimensional data set and curve graph, and every group of data are by target
The 3-dimensional amount of longitude and latitude and speed composition calculates separately the mean square error root of longitude, dimension and speed, analysis target tracking test
Effect.
The beneficial effects of the invention are that:
(1) present invention effectively can carry out real-time tracking to complicated nonlinear motion target, without to complexity
Moving target estimate motion model and complicated filtering algorithm to establish one；
(2) this invention address that the motion information using history adjusts neural network inner parameter, to realize to lower a period of time
Target estimation is carved, to simplify nonlinear filtering process；
(3) inner parameter of neural network is adjusted using back-propagation algorithm；Method using learning rate decaying reduces meter
Calculation amount simultaneously improves precision.
Detailed description of the invention
Fig. 1 is LSTM neural network structure schematic diagram；
Fig. 2 is neural network basic block diagram；
Fig. 3 (a) is GPS sensor needed for test；
Fig. 3 (b) is trolley tourist No. IV needed for test；
Fig. 4 is the method for tracking target flow chart based on LSTM neural network；
Fig. 5 (a) is to estimate test prognostic chart to amplified latitude to strong motion of automobile target following；
Fig. 5 (b) is to carry out estimation test prognostic chart to amplified longitude to strong motion of automobile target following；
Fig. 5 (c) is to carry out estimation test prognostic chart to speed to strong motion of automobile target following.
Specific embodiment
The present invention is further described with reference to the accompanying drawing.
Specific embodiment is as shown in figure 3, be a kind of target tracking algorism based on LSTM described in present embodiment, tool
Body process are as follows:
Step (1): acquiring the latitude and longitude information and velocity information of target, and the data of acquisition are carried out data processing；
It obtains first artificial cognition after the latitude and longitude information and velocity information of target and picks out wrong data, then will treated number
According to training set and test set is divided into, the weight and offset parameter of network structure are trained with training set, test set is used to detect mesh
Tracking accuracy is marked, and calculates target following square mean error amount.
Sensor acquires the survey of target longitude, latitude and speed under uniform motion, turning motion and the strong motion of automobile
Measure data.The detailed process of Measurement and Data Processing are as follows: amplify processing first firstly the need of pair warp and weft degree information, then need pair
Longitude and latitude carries out the normalized and data dissection process of data, to reduce calculation amount when training neural network parameter.
Data normalization process:
According to resolving:
In above-mentioned formula m indicate parsing data, x andThe mean value δ for respectively indicating true value and true value indicates true value
Mean square error.
Mean square error (MSE), expression formula is as follows:
In formula, yiThe correct option of i-th of data in first branch (batch) is represented, branch (batch) is that data exist
The number of every batch of training when participating in training, and y 'iFor predicted value.
Step (2): the LSTM neural network structure designed for monotrack；
The LSTM neural network structure of design is the neural network structure with one layer of hidden layer and multiple-input and multiple-output.For
Network is set to complicate and improve computational accuracy, therefore the concealed nodes number suitably chosen.Since network excessively complexity may
Lead to the tendency of over-fitting and increase calculation amount, therefore it is 10 that this step, which chooses concealed nodes number,.
Input layer: the dimension of input layer is 3, wherein being the longitude information, dimensional information and velocity information of target respectively.By
It is very small in the longitude and latitude data measured, therefore processing is amplified, in order to reduce calculation amount, and to amplified data
It is normalized and is parsed with data.
Hidden layer: as shown in Figure 1, LSTM neural network is that door is forgotten by three δ to control discarding or increase information, from
And realize the function of forgeing or remember, that is, forget door, input gate and out gate.In addition to these three doors, there are also input unit and
Output unit, door control unit is using sigmoid activation primitive, and input-output unit is using tanh activation primitive.
LSTM neural network is with traditional neural network the difference lies in that it is with timing and can use long historical information in short-term
As input.The number of concealed nodes is set as 10, and concealed nodes are unsuitable very few, and the complexity of the very few network of node is inadequate,
Node can excessively bring very big calculation amount.
Output layer: the dimension of output layer is 3, and activation primitive is linear function, exports longitude, dimension and the speed of target respectively
Spend information.
Step (3): LSTM neural network parameter is adjusted to realize target following；
Target tracking algorism process based on LSTM neural network substantially makees the 1st of training set to the 10th data
Information is inputted for history, calculates the 11st data；It is inputted again using the 2nd data of training set to the 11st data as history
Information calculates the 12nd data, parameter later and so on.By the data of calculating compared with truthful data, if error compared with
Greatly, then it adjusts neural network Inside Parameter Value to calculate again, until regulating optimized parameter.Test process: by the 1st of test set the
A to the 10th data as history input the 11st data of information prediction, using the 2nd of test set to the 11st data as
History inputs the 12nd data of information prediction, and prediction data later and so on finally calculates to draw out prediction curve
Root-mean-square error between prediction data and truthful data out.
LSTM Parameters of Neural Network Structure and test and data detection tracking effect are adjusted using training set data in step (3)
The specific implementation step of fruit is as follows:
Step (3.1) chronologically tieed up by 3 that the longitude and latitude and speed of target form by feed-in training set data, every group of data
Amount.
Step (3.2) sets initial parameter inside neural network, including sets the setting of 0.01, the number of iterations for learning rate
0.95 and neural network initial weight and offset parameter etc. are set as 1000, rate of decay.
Step (3.3) utilizes the target tracking algorism of LSTM neural network design by the 1st of training set to the 10th number
Information is inputted according to as history, calculates the 11st data；Again using the 2nd data of training set to the 11st data as history
Information is inputted, the 12nd data, parameter later and so on, the training set parameter predicted, relatively truer instruction are calculated
The parameter for practicing collection, to calculate the loss function in training set.
Whether the hundreds of step (3.4) the number of iterations occur carry, if executing step (3.5), otherwise skip step
(3.5), step (3.6) are executed.
Parameter is adjusted as new learning rate multiplied by after rate of decay in learning rate by step (3.5).
Step (3.6) uses the weight and offset parameter of Reverse Optimization Algorithm optimization neural network.
Whether step (3.7) reaches the number of iterations, if so then execute step (3.8), if not then jumping to step
(3.3)。
Step (3.8) chronologically feed-in test set data are inputted the 1st of test set to the 10th data as history
The 2nd of test set to the 11st data are inputted the 12nd data of information prediction by the 11st data of information prediction,
Prediction data later and so on.Obtaining output dimension is three-dimensional data set and curve graph, and every group of data are by target
The 3-dimensional amount of longitude and latitude and speed composition calculates separately the mean square error root of longitude, dimension and speed, analysis target tracking test
Effect.
Back-propagation algorithm:
Error of the loss function in output layer neuron:
Error of the loss function in middle layer: above formula obtains output layer error, according to the principle of error back propagation, can make
Current layer error is indicated with the compound function of one layer of all neuron error, and so on.Vector is expressed as:
Derivative of the loss function to weight:
Inverse of the loss function to bias term:
Back-propagation algorithm is substantially the application of chain type Rule for derivation, and Partial Variable can embody in such as Fig. 2 in above formula,
C is loss function in formula；It is from l-1 layers of k-th of neuron to the weight l j-th of neuron of layer；It is l
The bias term of j-th of neuron of layer；nlIt is the number of l layers of neuron；It is the weighting input of l j-th of neuron of layer；
It is the activation value of l j-th of neuron of layer, activation primitive is sigmoid function；It is loss function in j-th of nerve of l layer
The error of member, i.e.,
The expression formula of mean square error root (RMSE) is as follows:
Y and y' respectively indicates true value and predicted value in formula, and n indicates the quantity of test intensive data.
Embodiment:
For, based on the method for tracking target of LSTM, providing following examples described in above-mentioned specific embodiment:
The latitude and longitude information and velocity information of trolley in Fig. 3 (b) are obtained using the sensor in such as Fig. 3 (a).It is assumed that mesh
Be marked on movement over ground, when test robot it is motor-driven be it is random, the data of measurement be include uniform rectilinear's process, turned
Journey, accelerator and moderating process.It is 1Hz that rate is commented in the data acquisition of sensor, 1200 groups of data is measured in total, by data screening
Longitude, latitude and velocity information out.Using treated, data are put into the LSTM neural network model built as training set,
The number of iterations and learning rate are set, its voluntarily regulating networks inner parameter is made.300 groups of data are chosen from 1200 groups of data to make
The longitude and latitude of detection prediction and the error of true value are tested for test set, choose speed of 100 data as test set detection prediction
The error of degree and true value, draws out true value and predicted value curve, and calculates the mistake of the root mean square between true value and predicted value
Poor (RMSE), in this, as the evaluation index of target tracking accuracy
Test parameters under table the last 1 maneuver modeling
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art in the technical scope disclosed by the present invention, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
To compliance test result of the invention:
(a), (b), (c) are estimated the latitude, longitude and speed of strong maneuver modeling respectively in Fig. 5, blue in figure
Solid line represents true value, and red dotted line represents predicted value.By Test Drawing group and test data can be seen that prediction curve and
Real curve fitting preferably, shows that the method for tracking target based on LSTM is preferable to strong motor-driven target following effect.
Claims (4)
1. a kind of method for tracking target based on LSTM neural network characterized by comprising
(1) data of acquisition are carried out data processing by the latitude and longitude information and velocity information for acquiring target；
(2) it is designed for the LSTM neural network structure of monotrack；
(3) LSTM neural network parameter is adjusted to realize target following.
2. a kind of method for tracking target based on LSTM neural network according to claim 1, it is characterised in that: the step
Suddenly first manually method of discrimination picks out wrong data after the latitude and longitude information and velocity information of acquisition target in (1), then to warp
Latitude information amplifies processing first, then the normalized and data dissection process that need pair warp and weft degree to carry out data, then
By treated, data are divided into training set and test set, and the weight and offset parameter of network structure are trained with training set, test
Collection is used to detect target tracking accuracy, and calculates target following square mean error amount.
3. a kind of method for tracking target based on LSTM neural network according to claim 1, it is characterised in that: the step
Suddenly the LSTM neural network structure designed in (2) is the neural network structure with one layer of hidden layer and multiple-input and multiple-output；It is defeated
The dimension for entering layer is 3, wherein being the longitude information, dimensional information and velocity information of target respectively；The LSTM nerve net of hidden layer
Network is to forget doors by three δ to control discarding or increase information, to realize the function of forgetting or memory, in addition to forget door,
Outside input gate and out gate, there are also input units and output unit, and door control unit is using sigmoid activation primitive, input
Output unit is using tanh activation primitive；The number of concealed nodes is set as 10；The dimension of output layer is 3, activates letter
Number is linear function, exports longitude, dimension and the velocity information of target respectively.
4. a kind of method for tracking target based on LSTM neural network according to claim 1, it is characterised in that: the step
Suddenly (3) include,
(3.1) chronologically feed-in training set data, the 3-dimensional amount that every group of data are made of the longitude and latitude and speed of target；
(3.2) initial parameter inside setting neural network；
(3.3) using the method for tracking target of LSTM neural network design using the 1st of training set to the 10th data as going through
History inputs information, calculates the 11st data；The 2nd data of training set to the 11st data are inputted as history again and are believed
Breath, calculates the 12nd data, parameter later and so on, the training set parameter predicted, truer training set
Parameter, to calculate the loss function in training set；
(3.4) judge whether the hundreds of the number of iterations occur carry, if executing step (3.5), otherwise skip step
(3.5), step (3.6) are executed；
(3.5) parameter is adjusted as new learning rate multiplied by after rate of decay in learning rate；
(3.6) weight and offset parameter of Reverse optimization method optimization neural network are used；
(3.7) judge whether to reach the number of iterations, if so then execute step (3.8), if not then jumping to step (3.3)；
(3.8) the 1st of test set to the 10th data are inputted information prediction by chronologically feed-in test set data
The 2nd of test set to the 11st data are inputted the 12nd data of information prediction, later pre- by 11st data
Measured data and so on, obtaining output dimension is three-dimensional data set and curve graph, every group of data be by target longitude and latitude and
The 3-dimensional amount of speed composition, calculates separately the mean square error root of longitude, dimension and speed, analysis target tracking test effect.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041887.1A CN109740742A (en) | 2019-01-14 | 2019-01-14 | A kind of method for tracking target based on LSTM neural network |
| CA3067573A CA3067573A1 (en) | 2019-01-14 | 2020-01-13 | Target tracking systems and methods for uuv |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041887.1A CN109740742A (en) | 2019-01-14 | 2019-01-14 | A kind of method for tracking target based on LSTM neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109740742A true CN109740742A (en) | 2019-05-10 |
Family
ID=66365130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910041887.1A Pending CN109740742A (en) | 2019-01-14 | 2019-01-14 | A kind of method for tracking target based on LSTM neural network |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109740742A (en) |
| CA (1) | CA3067573A1 (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110212891A (en) * | 2019-06-02 | 2019-09-06 | 西安电子科技大学 | Non-linear filtering method based on NRP-LSTM |
| CN110309918A (en) * | 2019-07-05 | 2019-10-08 | 北京中科寒武纪科技有限公司 | Verification method, device and the computer equipment of Neural Network Online model |
| CN110443829A (en) * | 2019-08-05 | 2019-11-12 | 北京深醒科技有限公司 | It is a kind of that track algorithm is blocked based on motion feature and the anti-of similarity feature |
| CN110796093A (en) * | 2019-10-30 | 2020-02-14 | 上海眼控科技股份有限公司 | Target tracking method and device, computer equipment and storage medium |
| CN110794093A (en) * | 2019-11-11 | 2020-02-14 | 东北大学 | Precision compensation method for discharged caustic alkali concentration measuring device in evaporation process |
| CN110940971A (en) * | 2019-11-06 | 2020-03-31 | 四川川大智胜软件股份有限公司 | Radar target point trace recording method and device and storage medium |
| CN111027677A (en) * | 2019-12-02 | 2020-04-17 | 西安电子科技大学 | Multi-maneuvering-target tracking method based on depth certainty strategy gradient DDPG |
| CN111175719A (en) * | 2020-01-08 | 2020-05-19 | 中国船舶重工集团公司第七二四研究所 | Intelligent track starting method based on BP neural network |
| CN111241740A (en) * | 2020-01-25 | 2020-06-05 | 哈尔滨工程大学 | Fast and accurate calculation method for FPSO soft rigid arm stress |
| CN111445498A (en) * | 2020-03-19 | 2020-07-24 | 哈尔滨工程大学 | Target tracking method adopting Bi-L STM neural network |
| CN111552322A (en) * | 2020-04-29 | 2020-08-18 | 东南大学 | Unmanned aerial vehicle tracking method based on LSTM-particle filter coupling model |
| CN111931368A (en) * | 2020-08-03 | 2020-11-13 | 哈尔滨工程大学 | UUV target state estimation method based on GRU particle filter |
| CN112325877A (en) * | 2020-10-20 | 2021-02-05 | 北京嘀嘀无限科技发展有限公司 | Vehicle attribute determination method and device, storage medium and electronic equipment |
| CN113156320A (en) * | 2021-03-12 | 2021-07-23 | 山东大学 | Lithium ion battery SOC estimation method and system based on deep learning |
| CN113486960A (en) * | 2021-07-12 | 2021-10-08 | 安徽耀峰雷达科技有限公司 | Unmanned aerial vehicle tracking method and device based on long-time memory neural network, storage medium and computer equipment |
| CN117184382A (en) * | 2023-11-07 | 2023-12-08 | 成都诸元天成智能装备有限公司 | A stable system for submarine real-time gesture |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12012127B2 (en) * | 2019-10-26 | 2024-06-18 | Zoox, Inc. | Top-down view object detection and tracking |
| CN111932588B (en) * | 2020-08-07 | 2024-01-30 | 浙江大学 | Tracking method of airborne unmanned aerial vehicle multi-target tracking system based on deep learning |
| CN112083733B (en) * | 2020-08-19 | 2023-10-20 | 鹏城实验室 | Multi-underwater vehicle tracking control method based on data analysis |
| CN112130557B (en) * | 2020-08-19 | 2023-10-20 | 鹏城实验室 | Multi-underwater vehicle tracking control method, terminal and storage medium |
| CN112182954B (en) * | 2020-09-08 | 2023-05-23 | 上海大学 | LSTM-based fluid simulation data prediction model |
| CN112182961B (en) * | 2020-09-23 | 2023-01-31 | 中国南方电网有限责任公司超高压输电公司 | Converter station wireless network channel large-scale fading modeling prediction method |
| CN112613532B (en) * | 2020-11-26 | 2023-08-01 | 西安电子科技大学 | Moving target tracking method based on radar and cyclic neural network complement infrared fusion |
| CN112986940B (en) * | 2021-02-08 | 2023-04-18 | 北京无线电测量研究所 | Method for predicting radar sea clutter power in horizontal distance |
| CN113110430B (en) * | 2021-04-02 | 2024-01-30 | 大连海事大学 | Unmanned ship model-free fixed time accurate track tracking control method |
| CN113393032B (en) * | 2021-06-15 | 2023-09-12 | 西安电子科技大学 | Track circulation prediction method based on resampling |
| CN113703454B (en) * | 2021-08-26 | 2024-05-07 | 大连海事大学 | Four-degree-of-freedom-based unmanned ship power parameter real-time identification method |
| CN114034334B (en) * | 2021-09-15 | 2023-11-07 | 青岛理工大学 | Karst pipeline pollution source and flow identification method |
| CN114152257A (en) * | 2021-10-21 | 2022-03-08 | 南京航空航天大学 | Ship prediction navigation method based on attention mechanism and environment perception LSTM |
| CN114384509B (en) * | 2022-01-13 | 2024-04-26 | 桂林电子科技大学 | Safe driving decision generation method supported by intelligent driving vehicle data |
| CN114475502A (en) * | 2022-03-24 | 2022-05-13 | 中国第一汽车股份有限公司 | Method and device for protecting getting-off safety of passengers |
| CN114690222B (en) * | 2022-03-30 | 2023-08-08 | 杰能科世智能安全科技(杭州)有限公司 | Unmanned aerial vehicle detection track acquisition method based on wireless spectrum and unmanned aerial vehicle detection system |
| CN114862906B (en) * | 2022-04-11 | 2024-05-07 | 中山大学 | Visible light positioning and tracking method based on bidirectional cyclic convolutional neural network |
| CN116101464B (en) * | 2022-09-20 | 2024-07-02 | 西北工业大学深圳研究院 | Variable order fuzzy logic control method for unmanned underwater vehicle containing dynamic parameters |
| CN116300949A (en) * | 2023-03-29 | 2023-06-23 | 大连海事大学 | Course tracking control method and system for discrete time reinforcement learning unmanned ship |
| CN117633640A (en) * | 2023-11-23 | 2024-03-01 | 西安交通大学城市学院 | Unmanned aerial vehicle track prediction method based on flight mode identification |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108446324A (en) * | 2018-02-11 | 2018-08-24 | 重庆邮电大学 | A kind of GPS data reconstructing method based on long memory network LSTM in short-term |
| CN108509596A (en) * | 2018-04-02 | 2018-09-07 | 广州市申迪计算机***有限公司 | File classification method, device, computer equipment and storage medium |
| CN108682023A (en) * | 2018-03-21 | 2018-10-19 | 南京理工大学 | Close coupling Unscented kalman tracking filter algorithm based on Elman neural networks |
| CN108711847A (en) * | 2018-05-07 | 2018-10-26 | 国网山东省电力公司电力科学研究院 | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network |
| CN109001722A (en) * | 2018-05-30 | 2018-12-14 | 武汉理工大学 | A kind of ship track data fusion method based on LSTM model |
| CN109056056A (en) * | 2018-09-14 | 2018-12-21 | 西安理工大学 | A kind of Modelling of Crystal Growth in CZ-Si Pulling process thermal field identification Method |
-
2019
- 2019-01-14 CN CN201910041887.1A patent/CN109740742A/en active Pending
-
2020
- 2020-01-13 CA CA3067573A patent/CA3067573A1/en not_active Abandoned
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108446324A (en) * | 2018-02-11 | 2018-08-24 | 重庆邮电大学 | A kind of GPS data reconstructing method based on long memory network LSTM in short-term |
| CN108682023A (en) * | 2018-03-21 | 2018-10-19 | 南京理工大学 | Close coupling Unscented kalman tracking filter algorithm based on Elman neural networks |
| CN108509596A (en) * | 2018-04-02 | 2018-09-07 | 广州市申迪计算机***有限公司 | File classification method, device, computer equipment and storage medium |
| CN108711847A (en) * | 2018-05-07 | 2018-10-26 | 国网山东省电力公司电力科学研究院 | A kind of short-term wind power forecast method based on coding and decoding shot and long term memory network |
| CN109001722A (en) * | 2018-05-30 | 2018-12-14 | 武汉理工大学 | A kind of ship track data fusion method based on LSTM model |
| CN109056056A (en) * | 2018-09-14 | 2018-12-21 | 西安理工大学 | A kind of Modelling of Crystal Growth in CZ-Si Pulling process thermal field identification Method |
Non-Patent Citations (4)
| Title |
|---|
| ABHIK MILAN PAL 等: "Analytical Model for Determining the Sun’s Position at All Time Zones", 《INTERNATIONAL JOURNAL OF ENERGY ENGINEERING》 * |
| ZHIYUAN SHI 等: "LSTM-based Flight Trajectory Prediction", 《2018 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS》 * |
| 冯伟 等: "基于TensorFlow的俄语词汇标音***", 《计算机应用》 * |
| 陈慧岩 等: "《智能车辆理论与应用》", 31 July 2018 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110212891A (en) * | 2019-06-02 | 2019-09-06 | 西安电子科技大学 | Non-linear filtering method based on NRP-LSTM |
| CN110309918B (en) * | 2019-07-05 | 2020-12-18 | 安徽寒武纪信息科技有限公司 | Neural network online model verification method and device and computer equipment |
| CN110309918A (en) * | 2019-07-05 | 2019-10-08 | 北京中科寒武纪科技有限公司 | Verification method, device and the computer equipment of Neural Network Online model |
| CN110443829A (en) * | 2019-08-05 | 2019-11-12 | 北京深醒科技有限公司 | It is a kind of that track algorithm is blocked based on motion feature and the anti-of similarity feature |
| CN110796093A (en) * | 2019-10-30 | 2020-02-14 | 上海眼控科技股份有限公司 | Target tracking method and device, computer equipment and storage medium |
| CN110940971A (en) * | 2019-11-06 | 2020-03-31 | 四川川大智胜软件股份有限公司 | Radar target point trace recording method and device and storage medium |
| CN110794093A (en) * | 2019-11-11 | 2020-02-14 | 东北大学 | Precision compensation method for discharged caustic alkali concentration measuring device in evaporation process |
| US11893472B2 (en) | 2019-11-11 | 2024-02-06 | Northeastern University | Accuracy compensation method for discharge caustic alkali concentration measuring device in evaporation process |
| CN110794093B (en) * | 2019-11-11 | 2021-12-03 | 东北大学 | Precision compensation method for discharged caustic alkali concentration measuring device in evaporation process |
| CN111027677A (en) * | 2019-12-02 | 2020-04-17 | 西安电子科技大学 | Multi-maneuvering-target tracking method based on depth certainty strategy gradient DDPG |
| CN111175719A (en) * | 2020-01-08 | 2020-05-19 | 中国船舶重工集团公司第七二四研究所 | Intelligent track starting method based on BP neural network |
| CN111175719B (en) * | 2020-01-08 | 2023-10-17 | 中国船舶集团有限公司第七二四研究所 | BP neural network-based intelligent track initiation method |
| CN111241740B (en) * | 2020-01-25 | 2022-03-08 | 哈尔滨工程大学 | Fast and accurate calculation method for FPSO soft rigid arm stress |
| CN111241740A (en) * | 2020-01-25 | 2020-06-05 | 哈尔滨工程大学 | Fast and accurate calculation method for FPSO soft rigid arm stress |
| CN111445498A (en) * | 2020-03-19 | 2020-07-24 | 哈尔滨工程大学 | Target tracking method adopting Bi-L STM neural network |
| CN111552322A (en) * | 2020-04-29 | 2020-08-18 | 东南大学 | Unmanned aerial vehicle tracking method based on LSTM-particle filter coupling model |
| CN111931368A (en) * | 2020-08-03 | 2020-11-13 | 哈尔滨工程大学 | UUV target state estimation method based on GRU particle filter |
| CN112325877A (en) * | 2020-10-20 | 2021-02-05 | 北京嘀嘀无限科技发展有限公司 | Vehicle attribute determination method and device, storage medium and electronic equipment |
| CN112325877B (en) * | 2020-10-20 | 2022-09-02 | 北京嘀嘀无限科技发展有限公司 | Vehicle attribute determination method and device, storage medium and electronic equipment |
| CN113156320B (en) * | 2021-03-12 | 2023-05-30 | 山东大学 | Lithium ion battery SOC estimation method and system based on deep learning |
| CN113156320A (en) * | 2021-03-12 | 2021-07-23 | 山东大学 | Lithium ion battery SOC estimation method and system based on deep learning |
| CN113486960A (en) * | 2021-07-12 | 2021-10-08 | 安徽耀峰雷达科技有限公司 | Unmanned aerial vehicle tracking method and device based on long-time memory neural network, storage medium and computer equipment |
| CN117184382A (en) * | 2023-11-07 | 2023-12-08 | 成都诸元天成智能装备有限公司 | A stable system for submarine real-time gesture |
| CN117184382B (en) * | 2023-11-07 | 2024-01-19 | 成都诸元天成智能装备有限公司 | A stable system for submarine real-time gesture |
Also Published As
| Publication number | Publication date |
|---|---|
| CA3067573A1 (en) | 2020-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109740742A (en) | A kind of method for tracking target based on LSTM neural network | |
| Clark et al. | Vinet: Visual-inertial odometry as a sequence-to-sequence learning problem | |
| CN112766561B (en) | Attention mechanism-based generation type countermeasure track prediction method | |
| CN108621159B (en) | Robot dynamics modeling method based on deep learning | |
| Ko et al. | GP-UKF: Unscented Kalman filters with Gaussian process prediction and observation models | |
| CN109669049B (en) | Particle image velocity measurement method based on convolutional neural network | |
| CN108596327B (en) | Seismic velocity spectrum artificial intelligence picking method based on deep learning | |
| CN106548475A (en) | A kind of Forecasting Methodology of the target trajectory that spins suitable for space non-cooperative | |
| CN112356830A (en) | Intelligent parking method based on model reinforcement learning | |
| CN108115681A (en) | Learning by imitation method, apparatus, robot and the storage medium of robot | |
| CN107065890A (en) | A kind of unmanned vehicle intelligent barrier avoiding method and system | |
| CN108645413A (en) | The dynamic correcting method of positioning and map building while a kind of mobile robot | |
| CN110232412B (en) | Human gait prediction method based on multi-mode deep learning | |
| CN110728698A (en) | Multi-target tracking model based on composite cyclic neural network system | |
| CN109979161A (en) | A kind of tumble detection method for human body based on convolution loop neural network | |
| CN113447021B (en) | MEMS inertial navigation system positioning enhancement method based on LSTM neural network model | |
| CN105976397A (en) | Target tracking method based on half nonnegative optimization integration learning | |
| CN103839280B (en) | A kind of human body attitude tracking of view-based access control model information | |
| CN108226887A (en) | A kind of waterborne target rescue method for estimating state in the case of observed quantity transient loss | |
| CN113947616B (en) | Intelligent target tracking and loss rechecking method based on hierarchical perceptron | |
| CN107203271A (en) | Both hands recognition methods based on multi-sensor fusion technology | |
| CN109509207A (en) | The method that a kind of pair of point target and extension target carry out seamless tracking | |
| CN104715133B (en) | A kind of kinematics parameters in-orbit identification method and apparatus of object to be identified | |
| CN106485283A (en) | A kind of particle filter pedestrian target tracking based on Online Boosting | |
| CN114973417A (en) | Human body posture estimation method based on neural network and Kalman filtering |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190510 |