CN116977367A - Campus multi-target tracking method based on transform and Kalman filtering - Google Patents
Campus multi-target tracking method based on transform and Kalman filtering Download PDFInfo
- Publication number
- CN116977367A CN116977367A CN202310868814.6A CN202310868814A CN116977367A CN 116977367 A CN116977367 A CN 116977367A CN 202310868814 A CN202310868814 A CN 202310868814A CN 116977367 A CN116977367 A CN 116977367A
- Authority
- CN
- China
- Prior art keywords
- target tracking
- network
- network model
- target
- prediction
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000001914 filtration Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005259 measurement Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000000007 visual effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- G06T2207/20081—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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
A campus multi-target tracking method based on a transform and a Kalman filter comprises fusing a convolutional neural network and the transform model based on a PyTorch frame, and predicting by using the Kalman filter as a nonlinear tracker to generate a multi-target tracking network model; training a multi-target tracking network model; a multi-target tracking network model is detected. The invention combines the convolutional neural network, the transducer model and the Kalman filtering, improves the precision of the multi-target tracking task, and solves the problems of multi-target tracking target loss and the like in a target dense scene such as a campus. The method can be used for vehicle-road coordination, intelligent traffic, intelligent security, campus security, video monitoring and the like, and the multi-target tracking technology is widely applied in a plurality of important fields.
Description
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a campus multi-target tracking method based on a transform and Kalman filtering.
Background
The multi-target tracking technology is continuously applied to various aspects of production, life, military and the like at present, such as the fields of video monitoring, vehicle-road coordination, intelligent transportation, intelligent security, automatic driving and the like. Visual multi-target tracking is an important component of the computer field, but multi-target tracking in a complex background is still a very challenging task in the computer vision field due to various complex scenes such as occlusion, illumination, deformation, and the like.
The position of the object tracked by the visual multi-target has large change range, and the number of tracked objects is not fixed. In addition, the multi-target tracking problem typically tracks multiple objects of a given type, which have some apparent contour similarity. The following problems are generally solved in addition to factors such as object deformation and background interference: the automatic initialization and automatic termination of the tracking target, namely the problems of the appearance of a new target and the disappearance of an old target; when motion prediction and similarity judgment of a tracking target and interaction between occlusion of the tracking target and the target are carried out, the traditional method cannot quickly respond, so that tracking error is larger.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art, and provide the campus multi-target tracking method based on the transform and Kalman filtering, which has the advantages of simple method and high tracking precision.
The technical scheme adopted for solving the technical problems is as follows: a campus multi-target tracking method based on a transform and Kalman filtering comprises the following steps:
step 1, constructing a multi-target tracking network model
Based on a PyTorch frame, fusing a convolutional neural network and a transducer model, and predicting by using a Kalman filter as a nonlinear tracker to generate a multi-target tracking network model;
the multi-target tracking network model is formed by sequentially connecting a main network, a Transformer Encoder module, a Transformer Decoder module and a predictive head feedforward network, wherein the main network is a MobileNet V2 network and is used for extracting a characteristic diagram of an input picture to obtain a basic characteristic diagram; the Transformer Encoder module is used for adding position codes to the basic feature map output by the backbone network, and extracting global features of the basic feature map through a self-attention mechanism to obtain a global feature map; the Transformer Decoder module is used for reconstructing the global feature map output by the Transformer Encoder module and learning features, and obtaining token prediction after attention and mapping; the prediction head feedforward network is used for classifying and regressing;
the method for predicting by using the Kalman filter as the nonlinear tracker comprises the following steps:
a. according to the result output by the feedforward network of the pre-measuring head, the distance between targets is obtained by weighting the cosine distance of the appearance characteristic and the covariance distance between the pre-measuring state and the measuring state;
distance c between the targets i,j ：
c i,j ＝λd (1) +(1-λ)d (2)
d (1) ＝(d-y) T S -1 (d-y)
d (2) ＝min{1-cosα}
Wherein lambda is a hyper-parameter controlling the influence of each index on the merging cost, cos alpha is cosine similarity, d (2) For cosine distance, d is measurement distribution, y is prediction distribution, S is covariance matrix between measurement distribution and prediction distribution；
b. Obtaining a prediction result by using a Kalman filter according to the following formulas (1) and (2);
x z ＝Fx z-1 +B z u z (1)
in which x is z Is the predicted value of the state at the next moment, x z-1 Is the state at the current moment, F is the prediction step, B z To control the matrix, u z In order to control the vector quantity,for the prediction estimation error of the current moment, P z-1 For the prediction estimation error of the previous moment, F T For predicting the transpose of the step F matrix, Q is the covariance of the influence of external factors;
c. substituting the distance matrix between the targets into a cascade matching method and a IoU matching method to match the prediction result and the actual result of the Kalman filter, updating the state of the successfully matched target after the matching is successful, and establishing the unsuccessfully matched target as a new target to perform the next round of comparison matching;
d. updating the predicted track, and predicting the track of the next frame by using a Kalman filter;
step 2, training a multi-target tracking network model;
step 3, detecting the multi-target tracking network model
Setting the confidence threshold of the network parameter area of the multi-target tracking network model to be 0.05, wherein other parameters are network defaults, inputting the image in the test set into the trained multi-target tracking network model, and outputting a multi-target tracking result.
As a preferred technical solution, the method for adding position codes in the Transformer Encoder module in the step S1 is as follows: firstly, partitioning a basic feature map, and respectively carrying out horizontal direction position coding and vertical direction position coding according to the position of each picture block in an original map, wherein the horizontal direction position coding consists of sine values obtained according to the following formula (3) at even positions in the horizontal direction of the picture block and cosine values obtained according to the following formula (4) at odd positions;
wherein PE (p) β ,β) s Is of sine value, PE (p β ,β) c Is cosine value, p β For the position of a picture block in the basic feature map, beta is the index of the dimension of the picture block, d m An output dimension of Transformer Encoder module;
and splicing the horizontal direction position codes and the vertical direction position codes to obtain a vector with the same dimension as the output characteristic dimension of the convolutional neural network, and using the vector as the position code of the basic characteristic diagram.
As a preferred technical solution, in the step S1, the prediction head feedforward network includes a ReLU activation function and three linear layers with hidden layers, and the output of the last R layer of the prediction head feedforward network may be used as the conditional probability of each class, that is:
J＝softmax(Z (R) )
z in (R) Is the net input to the R layer and J is the R layer activity value.
As a preferable technical solution, the step S2 of training the multi-target tracking network model includes the following specific steps:
step 2.1. Setting training parameters
Drawing a frame from the MOT17-03-SDP data set every 0.1 second, acquiring the first 450 images as a training set and the last 50 images as a test set, uniformly adjusting image pixels in the training set and the test set to 1920×1080, setting the initial learning rate to lr to 0.0002, and optimizing the multi-target tracking network model by using an SGD (generalized gateway) optimizer;
s2.2. network model parameter initialization
Pre-training the backbone network by using an ImageNet data set to obtain weights and biases, wherein the weights and biases are used as initial weights and biases of the backbone network;
s2.3. training multi-target tracking network model
And inputting all images of the training set into the multi-target tracking network model for forward propagation and calculating a loss function until the convergence training of the loss function is finished, so as to obtain the trained multi-target tracking network model.
The beneficial effects of the invention are as follows:
the invention combines the convolutional neural network, the transducer model and the Kalman filtering, improves the precision of the multi-target tracking task, and solves the problems of multi-target tracking target loss and the like in a target dense scene such as a campus. The method can be used for vehicle-road coordination, intelligent traffic, intelligent security, campus security, video monitoring and the like, and the multi-target tracking technology is widely applied in a plurality of important fields.
Drawings
Fig. 1 is a schematic flow chart of a campus multi-target tracking method based on a transform and a kalman filter.
FIG. 2 is a schematic diagram of the architecture of the multi-objective tracking network model of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, but the present invention is not limited to the following embodiments.
In fig. 1, the campus multi-target tracking method based on transform and kalman filtering in this embodiment includes the following steps:
step 1, constructing a multi-target tracking network model
Based on a PyTorch framework, a convolutional neural network and a transducer model are fused, and a Kalman filter is used as a nonlinear tracker for prediction, so that a multi-target tracking network model is generated.
The multi-objective tracking network model is composed of a main network, a Transformer Encoder module, a Transformer Decoder module and a predictive head feedforward network which are sequentially connected in series, as shown in fig. 2.
The backbone network is a MobileNet V2 network and is used for extracting the feature map of the input picture to obtain a basic feature map.
Transformer Encoder module is configured to add position codes to the basic feature map output by the backbone network, and perform global feature extraction on the basic feature map through a self-attention mechanism, so as to obtain a global feature map.
The method for adding the position codes in the Transformer Encoder module comprises the following steps: firstly, partitioning a basic feature map, and respectively carrying out horizontal direction position coding and vertical direction position coding according to the position of each picture block in an original map, wherein the horizontal direction position coding consists of sine values obtained according to the following formula (1) at even positions in the horizontal direction of the picture block and cosine values obtained according to the following formula (2) at odd positions;
wherein PE (p) β ,β) s Is of sine value, PE (p β ,β) c Is cosine value, p β For the position of a picture block in the basic feature map, beta is the index of the dimension of the picture block, d m An output dimension of Transformer Encoder module;
and splicing the horizontal direction position codes and the vertical direction position codes to obtain a vector with the same dimension as the output characteristic dimension of the convolutional neural network, and using the vector as the position code of the basic characteristic diagram.
Transformer Decoder module for reconstructing the global feature map output by Transformer Encoder module and learning features, and obtaining token prediction after attention and mapping;
the pre-measurement head feedforward network comprises a ReLU activation function and three linear layers with hidden layers, wherein the ReLU activation function is used for normalizing center coordinates, heights and widths, the linear layers predict class labels through a softmax function, and the output of the last R layer of the pre-measurement head feedforward network can be used as the conditional probability of each class, namely:
J＝softmax(Z (R) )
z in (R) Is the net input to the R layer and J is the R layer activity value.
The method for predicting by using the Kalman filter as the nonlinear tracker comprises the following steps:
a. according to the result output by the feedforward network of the pre-measuring head, the distance between targets is obtained by weighting the cosine distance of the appearance characteristic and the covariance distance between the pre-measuring state and the measuring state;
wherein the distance c between the targets i,j ：
c i,j ＝λd (1) +(1-λ)d (2)
d (1) ＝(d-y) T S -1 (d-y)
d (2) ＝min{1-cosα}
Wherein lambda is a hyper-parameter controlling the influence of each index on the merging cost, cos alpha is cosine similarity, d (2) The cosine distance is d, the measurement distribution is y, the prediction distribution is y, and S is a covariance matrix between the measurement distribution and the prediction distribution;
b. obtaining a prediction result by using a Kalman filter according to the following formulas (3) and (4);
x z ＝Fx z-1 +B z u z (3)
in which x is z Is the predicted value of the state at the next moment, x z-1 Is the state at the current moment, F is the prediction step, B z To control the matrix, u z In order to control the vector quantity,for the prediction estimation error of the current moment, P z-1 To last onePrediction estimation error of time, F T For predicting the transpose of the step F matrix, Q is the covariance of the influence of external factors;
c. substituting the distance matrix between the targets into a cascade matching method and a IoU matching method to match the prediction result and the actual result of the Kalman filter, updating the state of the successfully matched target after the matching is successful, and establishing the unsuccessfully matched target as a new target to perform the next round of comparison matching;
d. updating the predicted track, and predicting the track of the next frame by using a Kalman filter;
step 2, training a multi-target tracking network model
Step 2.1. Setting training parameters
Drawing a frame from the MOT17-03-SDP data set every 0.1 second, acquiring the first 450 images as a training set and the last 50 images as a test set, uniformly adjusting image pixels in the training set and the test set to 1920×1080, setting the initial learning rate to lr to 0.0002, and optimizing the multi-target tracking network model by using an SGD (generalized gateway) optimizer;
step 2.2. Network model parameter initialization
Pre-training the backbone network by using an ImageNet data set to obtain weights and biases, wherein the weights and biases are used as initial weights and biases of the backbone network;
step 2.3 training a Multi-target tracking network model
Inputting all images of a training set into a multi-target tracking network model for forward propagation and calculating a loss function, predicting a normalized center coordinate through a pre-measurement head feedforward network, and obtaining a predicted class label by a linear layer through a softmax function, and performing next epoch training until the loss function convergence training is finished to obtain a trained multi-target tracking network model;
the softmax function is:
f(m,θ w，x )＝cos(m 1 ,θ w，x +m 3 )-m 2
wherein L is loss, k is category number, x is vector to be classified, s is scale super parameter, w is weight vector, p y Is the predicted posterior probability, g (·) function is used to mine difficult samples, i k To indicate a function, to dynamically specify whether a sample is misclassified,is the angle between w and x. />Is a marginal function, m in the formula 1 ≥1，m 2 >0，m 3 >0,h(t,θ w，x ,I k ) Is a weighted function of misclassified samples, where t is a preset hyper-parameter.
Step 3, detecting the multi-target tracking network model
Setting the confidence threshold of the network parameter area of the multi-target tracking network model to be 0.05, wherein other parameters are network defaults, inputting the image in the test set into the trained multi-target tracking network model, and outputting a multi-target tracking result.
Claims (3)
1. A campus multi-target tracking method based on a transform and Kalman filtering is characterized by comprising the following steps:
step 1, constructing a multi-target tracking network model
Based on a PyTorch frame, fusing a convolutional neural network and a transducer model, and predicting by using a Kalman filter as a nonlinear tracker to generate a multi-target tracking network model;
the multi-target tracking network model is formed by sequentially connecting a main network, a Transformer Encoder module, a Transformer Decoder module and a predictive head feedforward network in series, wherein the main network is a MobileNet V2 network and is used for extracting a characteristic diagram of an input picture to obtain a basic characteristic diagram; the Transformer Encoder module is used for adding position codes to the basic feature map output by the backbone network, and extracting global features of the basic feature map through a self-attention mechanism to obtain a global feature map; the Transformer Decoder module is used for reconstructing the global feature map output by the Transformer Encoder module and learning features, and obtaining token prediction after attention and mapping; the prediction head feedforward network is used for classifying and regressing;
the method for predicting by using the Kalman filter as the nonlinear tracker comprises the following steps:
a. according to the result output by the feedforward network of the pre-measuring head, the distance between targets is obtained by weighting the cosine distance of the appearance characteristic and the covariance distance between the pre-measuring state and the measuring state;
distance c between the targets i,j ：
c i,j ＝λd (1) +(1-λ)d (2)
d (1) ＝(d-y) T S -1 (d-y)
d (2) ＝min{1-cosα}
Wherein lambda is a hyper-parameter controlling the influence of each index on the merging cost, cos alpha is cosine similarity, d (2) The cosine distance is d, the measurement distribution is y, the prediction distribution is y, and S is a covariance matrix between the measurement distribution and the prediction distribution;
b. obtaining a prediction result by using a Kalman filter according to the following formulas (1) and (2);
x z ＝Fx z-1 +B z u z (1)
in which x is z Is the predicted value of the state at the next moment, x z-1 Is the state at the current moment, F is the prediction step, B z To control the matrix, u z In order to control the vector quantity,for the prediction estimation error of the current moment, P z-1 For the prediction estimation error of the previous moment, F T For predicting the transpose of the step F matrix, Q is the covariance of the influence of external factors;
c. substituting the distance matrix between the targets into a cascade matching method and a IoU matching method to carry out comparison matching on a prediction result and an actual result by using a Kalman filter, updating the state of the successfully matched target after the matching is successful, and establishing the unsuccessfully matched target as a new target to carry out next round of comparison matching;
d. updating the predicted track, and predicting the track of the next frame by using a Kalman filter;
step 2, training a multi-target tracking network model;
step 2.1. Setting training parameters
One frame of image was extracted every 0.1 second from the MOT17-03-SDP dataset, and the extracted image was extracted according to 9:1, dividing the model into a training set and a testing set, uniformly adjusting image pixels in the training set and the testing set to 1920 multiplied by 1080, setting the initial learning rate 1r to 0.0002, and optimizing the multi-target tracking network model by using an SGD (generalized gateway) optimizer;
step 2.2. Network model parameter initialization
Pre-training the backbone network by using an ImageNet data set to obtain weights and biases, wherein the weights and biases are used as initial weights and biases of the backbone network;
step 2.3 training a Multi-target tracking network model
Inputting all images of the training set into the multi-target tracking network model for forward propagation and calculating a loss function until the convergence training of the loss function is finished, so as to obtain a trained multi-target tracking network model;
step 3, detecting the multi-target tracking network model
Setting the confidence threshold of the network parameter area of the multi-target tracking network model to be 0.05, wherein other parameters are network defaults, inputting the image in the test set into the trained multi-target tracking network model, and outputting a multi-target tracking result.
2. The campus multi-target tracking method based on the transform and kalman filtering according to claim 1, wherein the method for adding the position code in the Transformer Encoder module in the step S1 is as follows: firstly, partitioning a basic feature map, and respectively carrying out horizontal direction position coding and vertical direction position coding according to the position of each picture block in an original map, wherein the horizontal direction position coding consists of sine values obtained according to the following formula (3) at even positions in the horizontal direction of the picture block and cosine values obtained according to the following formula (4) at odd positions;
wherein PE (p) β ,β) s Is of sine value, PE (p β ,β) c Is cosine value, p β For the position of a picture block in the basic feature map, beta is the index of the dimension of the picture block, d m An output dimension of Transformer Encoder module;
and splicing the horizontal direction position codes and the vertical direction position codes to obtain a vector with the same dimension as the output characteristic dimension of the convolutional neural network, and using the vector as the position code of the basic characteristic diagram.
3. The campus multi-objective tracking method based on the transform and kalman filtering according to claim 1, wherein the predicting the head feed-forward network in step S1 includes a ReLU activation function and three linear layers with hidden layers, and the output of the last R layer of the predicting head feed-forward network can be used as the conditional probability of each class, namely:
J＝softmax(Z (R) )
z in (R) Is the net input to the R layer and J is the R layer activity value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310868814.6A CN116977367A (en) | 2023-07-14 | 2023-07-14 | Campus multi-target tracking method based on transform and Kalman filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310868814.6A CN116977367A (en) | 2023-07-14 | 2023-07-14 | Campus multi-target tracking method based on transform and Kalman filtering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116977367A true CN116977367A (en) | 2023-10-31 |
Family
ID=88477638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310868814.6A Pending CN116977367A (en) | 2023-07-14 | 2023-07-14 | Campus multi-target tracking method based on transform and Kalman filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116977367A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117241133A (en) * | 2023-11-13 | 2023-12-15 | 武汉益模科技股份有限公司 | Visual work reporting method and system for multi-task simultaneous operation based on non-fixed position |
| CN117788511A (en) * | 2023-12-26 | 2024-03-29 | 兰州理工大学 | Multi-expansion target tracking method based on deep neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10269125B1 (en) * | 2018-10-05 | 2019-04-23 | StradVision, Inc. | Method for tracking object by using convolutional neural network including tracking network and computing device using the same |
| CN113674328A (en) * | 2021-07-14 | 2021-11-19 | 南京邮电大学 | Multi-target vehicle tracking method |
| CN114529799A (en) * | 2022-01-06 | 2022-05-24 | 浙江工业大学 | Aircraft multi-target tracking method based on improved YOLOV5 algorithm |
| CN115861884A (en) * | 2022-12-06 | 2023-03-28 | 中南大学 | Video multi-target tracking method, system, device and medium in complex scene |
| CN116245916A (en) * | 2023-05-11 | 2023-06-09 | 中国人民解放军国防科技大学 | Unmanned ship-oriented infrared ship target tracking method and device |
-
2023
- 2023-07-14 CN CN202310868814.6A patent/CN116977367A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10269125B1 (en) * | 2018-10-05 | 2019-04-23 | StradVision, Inc. | Method for tracking object by using convolutional neural network including tracking network and computing device using the same |
| CN113674328A (en) * | 2021-07-14 | 2021-11-19 | 南京邮电大学 | Multi-target vehicle tracking method |
| CN114529799A (en) * | 2022-01-06 | 2022-05-24 | 浙江工业大学 | Aircraft multi-target tracking method based on improved YOLOV5 algorithm |
| CN115861884A (en) * | 2022-12-06 | 2023-03-28 | 中南大学 | Video multi-target tracking method, system, device and medium in complex scene |
| CN116245916A (en) * | 2023-05-11 | 2023-06-09 | 中国人民解放军国防科技大学 | Unmanned ship-oriented infrared ship target tracking method and device |
Non-Patent Citations (1)
| Title |
|---|
| 王利等: "基于双解码器的Transformer多目标跟踪方法", 计算机应用》, vol. 43, no. 6, 10 June 2023 (2023-06-10), pages 2 - 3 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117241133A (en) * | 2023-11-13 | 2023-12-15 | 武汉益模科技股份有限公司 | Visual work reporting method and system for multi-task simultaneous operation based on non-fixed position |
| CN117241133B (en) * | 2023-11-13 | 2024-02-06 | 武汉益模科技股份有限公司 | Visual work reporting method and system for multi-task simultaneous operation based on non-fixed position |
| CN117788511A (en) * | 2023-12-26 | 2024-03-29 | 兰州理工大学 | Multi-expansion target tracking method based on deep neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110070074B (en) | Method for constructing pedestrian detection model | |
| CN116977367A (en) | Campus multi-target tracking method based on transform and Kalman filtering | |
| CN108573496B (en) | Multi-target tracking method based on LSTM network and deep reinforcement learning | |
| CN108470355B (en) | Target tracking method fusing convolution network characteristics and discriminant correlation filter | |
| CN114972418B (en) | Maneuvering multi-target tracking method based on combination of kernel adaptive filtering and YOLOX detection | |
| CN112836640B (en) | Single-camera multi-target pedestrian tracking method | |
| CN111161315B (en) | Multi-target tracking method and system based on graph neural network | |
| CN111582349B (en) | Improved target tracking algorithm based on YOLOv3 and kernel correlation filtering | |
| CN112348849A (en) | Twin network video target tracking method and device | |
| CN111932583A (en) | Space-time information integrated intelligent tracking method based on complex background | |
| CN110728698B (en) | Multi-target tracking system based on composite cyclic neural network system | |
| CN110660083A (en) | Multi-target tracking method combined with video scene feature perception | |
| CN111080675A (en) | Target tracking method based on space-time constraint correlation filtering | |
| CN109145836A (en) | Ship target video detection method based on deep learning network and Kalman filtering | |
| CN109886994A (en) | Adaptive sheltering detection system and method in video tracking | |
| CN108320306A (en) | Merge the video target tracking method of TLD and KCF | |
| CN111862145A (en) | Target tracking method based on multi-scale pedestrian detection | |
| CN110310305A (en) | A kind of method for tracking target and device based on BSSD detection and Kalman filtering | |
| CN114972439A (en) | Novel target tracking algorithm for unmanned aerial vehicle | |
| CN117576379B (en) | Target detection method based on meta-learning combined attention mechanism network model | |
| CN113536939B (en) | Video duplication removing method based on 3D convolutional neural network | |
| CN108320301B (en) | Target tracking optimization method based on tracking learning detection | |
| CN113129336A (en) | End-to-end multi-vehicle tracking method, system and computer readable medium | |
| CN110111358B (en) | Target tracking method based on multilayer time sequence filtering | |
| CN116245913A (en) | Multi-target tracking method based on hierarchical context guidance |
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 |