CN110188753A - One kind being based on dense connection convolutional neural networks target tracking algorism - Google Patents
One kind being based on dense connection convolutional neural networks target tracking algorism Download PDFInfo
- Publication number
- CN110188753A CN110188753A CN201910425297.9A CN201910425297A CN110188753A CN 110188753 A CN110188753 A CN 110188753A CN 201910425297 A CN201910425297 A CN 201910425297A CN 110188753 A CN110188753 A CN 110188753A
- Authority
- CN
- China
- Prior art keywords
- characteristic pattern
- network
- neural networks
- convolutional neural
- image
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 29
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 42
- 238000000605 extraction Methods 0.000 claims abstract description 30
- 239000000284 extract Substances 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims description 13
- 230000004044 response Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 description 15
- 238000001514 detection method Methods 0.000 description 10
- 230000013016 learning Effects 0.000 description 7
- 238000000034 method Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000004064 recycling Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Classifications
-
- 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
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses one kind to be based on dense connection convolutional neural networks target tracking algorism, the following steps are included: S101, extraction input picture feature: in advance by using dense connection convolutional neural networks as feature extraction network, obtaining and extract output characteristics of image；S103, output characteristic pattern processing: bilinear interpolation is carried out to extraction output characteristics of image is obtained, obtains the characteristic pattern of bilinear interpolation；S105, acquisition characteristic pattern is calculated.The utility model has the advantages that the present invention replaces the feature extraction network of the twin convolutional network of tradition to make network obtain stronger ability in feature extraction using the convolutional neural networks of dense connection, make bilinear interpolation by the characteristic pattern exported to feature extraction network, improve characteristic pattern resolution ratio, improve tracing algorithm positioning accuracy, in addition RPN layers are added on the basis of traditional twin convolutional network, enhance track algorithm to the separating capacity of target and background.
Description
Technical field
The present invention relates to deep learnings and technical field of computer vision, it particularly relates to which a kind of be based on dense connection
Convolutional neural networks target tracking algorism.
Background technique
Target following technology is an important branch in Computer Vision Task, is widely used in automatic Pilot, depending on
The fields such as frequency monitoring and robot.Conventional target track algorithm, which depends on, marks feature and correlation filtering (such as by hand
KCF, TLD etc.), there is higher frame per second, but accuracy rate and robustness are lower, it is difficult to meet application request.In recent years
With the rise of artificial intelligence and deep learning, convolutional neural networks algorithm steps into target tracking domain, and achieves
Original performance and achievement, wherein algorithm frame based on twin convolutional network is by its good performance and succinct
Network structure, international computer vision top meeting in recent years and tracking race in receive very big concern.
RPN (region proposal network) layer is proposed in algorithm of target detection faster-rcnn earliest
, it is instead of traditional candidate region (region proposal) generating algorithm (such as sliding window), for generating detection
Candidate region needed for algorithm.Its main feature is that operation efficiency height does not have redundant computation, the candidate region of high quality, pole can be generated
Big improves the performance of related two-stage detection algorithm.
And the convolutional neural networks (densenet) intensively connected are the novel volumes that the best paper of CVPR20117 is proposed
Product neural network structure, has inwardly used for reference resnet, and has carried out innovative improvement, and network structure is simple, but very high
Effect, densenet is using multiplexing and transmitting of the convolutional network to feature of the intensive connection reinforcement between convolutional layer, to solve
Many pain spots of deep layer network, such as gradient disperse, parameter amount is larger etc..The characteristic pattern obtained by feature extraction network is usual
There is the semantic information of higher level, but resolution ratio is smaller, the original location information of image is lost seriously, and target positioning accurate is influenced
Degree.
For the problems in the relevant technologies, currently no effective solution has been proposed.
Summary of the invention
For the problems in the relevant technologies, the present invention proposes a kind of based on dense connection convolutional neural networks target following calculation
Method, by replacing traditional alexnet as feature extraction using densenet on the basis of the algorithm of the twin convolutional network of tradition
Network is added RPN layers simultaneously, further improves track algorithm to the separating capacity of target and background, while tradition tracking being appointed
Business is converted to the operation efficiency that single sample learning (one-shot) Detection task improves the algorithm reasoning stage, to overcome existing phase
Above-mentioned technical problem present in the technology of pass.
The technical scheme of the present invention is realized as follows:
One kind being based on dense connection convolutional neural networks target tracking algorism, comprising the following steps:
S101, input picture feature is extracted: in advance by using dense connection convolutional neural networks as feature extraction net
Network obtains and extracts output characteristics of image；
S103, output characteristic pattern processing: bilinear interpolation is carried out to extraction output characteristics of image is obtained, bilinearity is obtained and inserts
The characteristic pattern of value；
S105, acquisition characteristic pattern is calculated；Will acquire bilinear interpolation characteristic pattern input respectively RPN layers classification branch and
It returns branch and tracking target image and tracing area image is filtered and distinguished to the output of two branches, when algorithm exports
It is coordinate of the target response value highest point on characteristic pattern as final characteristic pattern.
Further, image characteristics extraction is obtained using following formula in the step S101:
Xe=He ([X0,X1......Xe-1,]), wherein；
He is that network extracts characteristic operation, ([X0,X1......Xe-1,]) and it is to make the characteristic pattern of first layer to the last layer
Channel merges, and Xe is characterized the output for extracting network.
Further, the step S101 output characteristics of image can be expressed asWithWherein；
Z is target image, and X is tracing area image,Convolutional neural networks intensively to connect extract characteristic operation.
Further, describedWithThe classification branch of RPN layers of input is carried out respectively with recurrence branch and to two
Road output is filtered, and is obtained using following formula:
Wherein, A is that RPN exports characteristic pattern, the width and height of w and h flag output characteristic pattern, and k is characterized each pixel in figure
The number of the boundingbox of the different scale and length-width ratio that are mapped in original image, * are correlation filtering operation.
Beneficial effects of the present invention:
1, the present invention promotes the ability in feature extraction of network by introducing densenet, is replaced using densenet
Feature extraction network of the alexnet as twin network, densenet are a kind of with the convolutional neural networks intensively connected,
Main thought is to connect all layers under the premise of maximum information is transmitted between layers in guaranteeing network, often
Output of one layer of the input from all layers of front, the output of current layer also can be as the input of each convolutional layer later, networks
Dense connection between each layer not only reduces the calculation amount of each layer, realizes the recycling of feature, and strengthens feature
It propagates, alleviates the gradient diffusing phenomenon of deep layer network, further improve network performance.
2, the present invention improves the resolution ratio of characteristic pattern by interpolation algorithm, restores the position letter lost in characteristic extraction procedure
Breath, to improve the positioning accuracy of algorithm.
3, the present invention is by joined RPN layer on the basis of conventional twin network trace algorithm, to track target and background into
Further classification and recurrence are gone, introduced RPN layer is by supervised learning part and correlation filtering part structure through the invention
At it is that preceding background class and target frame return that supervised learning part is made of recurrence branch and classification branch, correlation filtering portion
Divide and be responsible for carrying out the characteristic pattern of target and tracing area in correlation filtering operation acquisition output response figure, while the present invention will be traditional
Tracing task is converted for single pattern detection task, and efficiency of algorithm is improved.
In conclusion the present invention replaces the feature of the twin convolutional network of tradition to mention using the convolutional neural networks of dense connection
Network is taken to make network obtain stronger ability in feature extraction, is made by the characteristic pattern exported to feature extraction network double
Linear interpolation improves characteristic pattern resolution ratio, improves tracing algorithm positioning accuracy, in addition on the basis of traditional twin convolutional network
RPN layers are added, enhances track algorithm to the separating capacity of target and background.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without creative efforts, can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is a kind of process based on dense connection convolutional neural networks target tracking algorism according to an embodiment of the present invention
Schematic diagram；
Fig. 2 is a kind of RPN based on dense connection convolutional neural networks target tracking algorism according to an embodiment of the present invention
Scene application drawing；
Fig. 3 is according to an embodiment of the present invention a kind of based on dense connection convolutional neural networks target tracking algorism
Densenet scene application drawing；
Fig. 4 is according to an embodiment of the present invention a kind of based on dense connection convolutional neural networks target tracking algorism
Dense block scene application drawing.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art's every other embodiment obtained belong to what the present invention protected
Range.
According to an embodiment of the invention, providing a kind of based on dense connection convolutional neural networks target tracking algorism.
As shown in Figs 1-4, according to an embodiment of the present invention to be based on dense connection convolutional neural networks target tracking algorism, packet
Include following steps:
S101, input picture feature is extracted: in advance by using dense connection convolutional neural networks as feature extraction net
Network obtains and extracts output characteristics of image；
S103, output characteristic pattern processing: bilinear interpolation is carried out to extraction output characteristics of image is obtained, bilinearity is obtained and inserts
The characteristic pattern of value；
S105, acquisition characteristic pattern is calculated；Will acquire bilinear interpolation characteristic pattern input respectively RPN layers classification branch and
It returns branch and tracking target image and tracing area image is filtered and distinguished to the output of two branches, when algorithm exports
It is coordinate of the target response value highest point on characteristic pattern as final characteristic pattern.
With the aid of the technical scheme, it by promoting the ability in feature extraction of network by introducing densenet, uses
Densenet replaces feature extraction network of the alexnet as twin network, and densenet is a kind of with the volume intensively connected
Product neural network, main thought are will own under the premise of maximum information is transmitted between layers in guaranteeing network
Layer connects, the output of each layer of input from all layers of front, the output of current layer also can as after each convolution
The input of layer, the dense connection between each layer of network not only reduce the calculation amount of each layer, realize the recycling of feature, and
And feature propagation is strengthened, the gradient diffusing phenomenon of deep layer network is alleviated, network performance is further improved.And pass through interpolation
Algorithm improves the resolution ratio of characteristic pattern, the location information lost in characteristic extraction procedure is restored, to improve the positioning accurate of algorithm
Degree.It joined RPN layers additionally by the basis of conventional twin network trace algorithm, tracking target and background carried out further
Classification and recurrence, introduced RPN layer is made of supervised learning part and correlation filtering part through the invention, supervised learning
Part is made of recurrence branch and classification branch, is that preceding background class and target frame return, correlation filtering part is responsible for mesh
The characteristic pattern of mark and tracing area carries out correlation filtering operation and obtains output response figure, while the present invention turns traditional tracing task
It has been changed to single pattern detection task, has improved efficiency of algorithm.
In addition, in one embodiment, image characteristics extraction is obtained using following formula in the step S101:
Xe=He ([X0, X1......Xe-1,]), wherein；
He is that network extracts characteristic operation, ([X0, X1......Xe-1,]) and it is to make the characteristic pattern of first layer to the last layer
Channel merges, and Xe is characterized the output for extracting network.
In addition, in one embodiment, the step S101 output characteristics of image can be expressed asWith
Wherein；
Z is target image, and X is tracing area image,Convolutional neural networks intensively to connect extract characteristic operation.
In addition, in one embodiment, it is describedWithClassification branch and the recurrence of RPN layers of input are carried out respectively
Branch is simultaneously filtered the output of two branches, is obtained using following formula:
Wherein, A is that RPN exports characteristic pattern, the width and height of w and h flag output characteristic pattern, and k is characterized each pixel in figure
The number of the boundingbox of the different scale and length-width ratio that are mapped in original image, * are correlation filtering operation.
In addition, in one embodiment, for above-mentioned bilinear interpolation, its core concept of bilinear interpolation tangible two
Once linear interpolation is carried out on a direction respectively, such as expects unknown function f in the value of point P=(x, y), it is assumed that known function
F is in Q11=(x1, y1), Q12=(x1, y2), Q21=(x2, y1), Q22=(x2, y2) four points value.
It carries out linear interpolation in the x direction first, obtains:
Then linear interpolation is carried out on the direction y again, obtained:
Then the approximation of point f (x, y) is just obtained
In addition, in one embodiment, for above-mentioned recurrence branch (reg) and classification branch (cls), due to being added
RPN layers, so use multitask loss function as the loss function of this algorithm, wherein classification branch use cross entropy as
Loss function returns branch and uses smooth L1loss as loss function.Algorithm training process can be divided into following steps:
Step S1: assuming that Ax, Ay, Aw, Ah indicate the centre coordinate and wide high, Tx, Ty of anchor, Tw, Th expression
The centre coordinate of ground truth and wide height.Therefore normalized cumulant can be indicated by following mathematic(al) representations:
Step S2: the mathematic(al) representation of known smooth L1loss are as follows:
Branch rood loss function must be returned by training step S1 are as follows:
Step S3: by above-mentioned steps S1, S2 and multitask loss function define available of the invention obtain and finally lose
Function are as follows:
Loss=Lcls+λLreg
Wherein Lcls is classification branch rood cross entropy loss function, and Lreg, which is that step S2 is described, obtains smooth L1 loss
Function, lambda parameter are hyper parameter for balanced sort and return two rood weights.
In addition to this, traditional tracing task can be converted to single sample (one-shot) detection by following steps derivation to appoint
Business.
Step S1, it is assumed that algorithm uses loss function of the average loss as network, therefore the main target of training process
Weight matrix W is exactly found, following mathematic(al) representation is set up:
Wherein, xi is i-th of input sample, ψ (xi；W) indicate that anticipation function, li indicate the actual value of i-th of sample.Step
Rapid S2, it will be assumed that function ω is feedforward equation, by (z；W it) is mapped as W, therefore the mathematic(al) representation in step S1 can indicate
Are as follows:
Wherein, zi is i-th of template sample, i.e. target area image.
Step S3, it is assumed that function isTwin network extracts characteristic operation, and ζ is RPN layers of operation, therefore mathematics in S step 2
Expression formula can convert are as follows:
In addition, in one embodiment, for the RPN (region proposal network), RPN layers of production
During raw candidate region, using a variety of policy selection candidate regions as abandoned the candidate region of range image center too far,
Cosine Window penalty term is added to candidate region again ranking and non-maxima suppression (NMS) etc..In the algorithm forward inference stage,
Target area in first frame is admitted to template branch and generates cls, reg convolution kernel, later template branch
It is removed, residue frame passes sequentially through detection branch and cls, and reg convolution kernel, which calculates, generates final characteristic pattern.
In addition, in one embodiment, implement this algorithm on VOT2015 data set, and with other state-of-the-
The track algorithm (track algorithm including tradition based on twin network) of art compares performance indexes, as a result as shown in the table:
In addition, in one embodiment, implement this algorithm on VOT2016 data set, and with other state-of-the-
The track algorithm (track algorithm including tradition based on twin network) of art compares performance indexes, as a result as shown in the table:
In conclusion by means of above-mentioned technical proposal of the invention, it can be achieved that following effect:
1, the present invention promotes the ability in feature extraction of network by introducing densenet, is replaced using densenet
Feature extraction network of the alexnet as twin network, densenet are a kind of with the convolutional neural networks intensively connected,
Main thought is to connect all layers under the premise of maximum information is transmitted between layers in guaranteeing network, often
Output of one layer of the input from all layers of front, the output of current layer also can be as the input of each convolutional layer later, networks
Dense connection between each layer not only reduces the calculation amount of each layer, realizes the recycling of feature, and strengthens feature
It propagates, alleviates the gradient diffusing phenomenon of deep layer network, further improve network performance.
2, the present invention improves the resolution ratio of characteristic pattern by interpolation algorithm, restores the position letter lost in characteristic extraction procedure
Breath, to improve the positioning accuracy of algorithm.
3, the present invention is by joined RPN layer on the basis of conventional twin network trace algorithm, to track target and background into
Further classification and recurrence are gone, introduced RPN layer is by supervised learning part and correlation filtering part structure through the invention
At it is that preceding background class and target frame return that supervised learning part is made of recurrence branch and classification branch, correlation filtering portion
Divide and be responsible for carrying out the characteristic pattern of target and tracing area in correlation filtering operation acquisition output response figure, while the present invention will be traditional
Tracing task is converted for single pattern detection task, and efficiency of algorithm is improved.
In conclusion the present invention uses densenet to replace alexnet as the feature extraction network of twin convolutional network, mention
The high performance of network, while bilinear interpolation has been carried out to the characteristic pattern of output, the resolution ratio of characteristic pattern is improved, sufficiently
The information in characteristic pattern is utilized.And RPN layers are increased on the basis of twin convolutional network structure, is increasing track algorithm
While to the separating capacity of target and background, traditional tracing task is converted into single sample (one-shot) Detection task,
Improving the operational efficiency of algorithm, accuracy and speed has great promotion compared to traditional track algorithm.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (4)
1. one kind is based on dense connection convolutional neural networks target tracking algorism, which comprises the following steps:
S101, it extracts input picture feature: being obtained in advance by using dense connection convolutional neural networks as feature extraction network
Take extraction output characteristics of image；
S103, output characteristic pattern processing: bilinear interpolation is carried out to extraction output characteristics of image is obtained, obtains bilinear interpolation
Characteristic pattern；
S105, acquisition characteristic pattern is calculated；The characteristic pattern that will acquire bilinear interpolation inputs RPN layers of classification branch and recurrence respectively
Branch is simultaneously filtered to the output of two branches and distinguishes tracking target image and tracing area image, when algorithm output is mesh
Coordinate of the response highest point on characteristic pattern is marked as final characteristic pattern.
2. according to claim 1 be based on dense connection convolutional neural networks target tracking algorism, which is characterized in that described
Image characteristics extraction is obtained using following formula in step S101:
Xe=He ([X0,X1......Xe-1,]), wherein；
He is that network extracts characteristic operation, ([X0,X1......Xe-1,]) and it is that the characteristic pattern of first layer to the last layer is made into channel
Merge, Xe is characterized the output for extracting network.
3. according to claim 1 be based on dense connection convolutional neural networks target tracking algorism, which is characterized in that described
Step S101 output characteristics of image can be expressed asWithWherein；
Z is target image, and X is tracing area image,Convolutional neural networks intensively to connect extract characteristic operation.
4. according to claim 3 be based on dense connection convolutional neural networks target tracking algorism, which is characterized in that describedWithThe classification branch of RPN layers of input is carried out respectively and returns branch and the output of two branches is filtered, and is adopted
It is obtained with following formula:
Wherein, A is that RPN exports characteristic pattern, the width and height of w and h flag output characteristic pattern, and k is characterized each pixel-map in figure
The number of the boundingbox of different scale and length-width ratio into original image, * are correlation filtering operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910425297.9A CN110188753A (en) | 2019-05-21 | 2019-05-21 | One kind being based on dense connection convolutional neural networks target tracking algorism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910425297.9A CN110188753A (en) | 2019-05-21 | 2019-05-21 | One kind being based on dense connection convolutional neural networks target tracking algorism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110188753A true CN110188753A (en) | 2019-08-30 |
Family
ID=67717107
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910425297.9A Pending CN110188753A (en) | 2019-05-21 | 2019-05-21 | One kind being based on dense connection convolutional neural networks target tracking algorism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110188753A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689556A (en) * | 2019-09-09 | 2020-01-14 | 苏州臻迪智能科技有限公司 | Tracking method and device and intelligent equipment |
| CN110689559A (en) * | 2019-09-30 | 2020-01-14 | 长安大学 | Visual target tracking method based on dense convolutional network characteristics |
| CN111091585A (en) * | 2020-03-19 | 2020-05-01 | 腾讯科技（深圳）有限公司 | Target tracking method, device and storage medium |
| CN111415373A (en) * | 2020-03-20 | 2020-07-14 | 北京以萨技术股份有限公司 | Target tracking and segmenting method, system and medium based on twin convolutional network |
| CN111724409A (en) * | 2020-05-18 | 2020-09-29 | 浙江工业大学 | Target tracking method based on densely connected twin neural network |
| CN112749602A (en) * | 2019-10-31 | 2021-05-04 | 北京市商汤科技开发有限公司 | Target query method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182388A (en) * | 2017-12-14 | 2018-06-19 | 哈尔滨工业大学（威海） | A kind of motion target tracking method based on image |
| CN109035558A (en) * | 2018-06-12 | 2018-12-18 | 武汉市哈哈便利科技有限公司 | A kind of commodity recognizer on-line study system for self-service cabinet |
| CN109410251A (en) * | 2018-11-19 | 2019-03-01 | 南京邮电大学 | Method for tracking target based on dense connection convolutional network |
| CN109726657A (en) * | 2018-12-21 | 2019-05-07 | 万达信息股份有限公司 | A kind of deep learning scene text recognition sequence method |
-
2019
- 2019-05-21 CN CN201910425297.9A patent/CN110188753A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182388A (en) * | 2017-12-14 | 2018-06-19 | 哈尔滨工业大学（威海） | A kind of motion target tracking method based on image |
| CN109035558A (en) * | 2018-06-12 | 2018-12-18 | 武汉市哈哈便利科技有限公司 | A kind of commodity recognizer on-line study system for self-service cabinet |
| CN109410251A (en) * | 2018-11-19 | 2019-03-01 | 南京邮电大学 | Method for tracking target based on dense connection convolutional network |
| CN109726657A (en) * | 2018-12-21 | 2019-05-07 | 万达信息股份有限公司 | A kind of deep learning scene text recognition sequence method |
Non-Patent Citations (5)
| Title |
|---|
| BO LI 等: "High Performance Visual Tracking with Siamese Region Proposal Network", 《2018 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| BO LI 等: "SiamRPN++: Evolution of Siamese Visual Tracking with Very Deep Networks", 《ARXIV》 * |
| GAO HUANG 等: "Densely Connected Convolutional Networks", 《ARXIV》 * |
| MOHAMED H. ABDELPAKEY 等: "DensSiam: End-to-End Densely-Siamese Network with Self-Attention Model for Object Tracking", 《ARXIV》 * |
| 郭智 等: "基于深度卷积神经网络的遥感图像飞机目标检测方法", 《电子与信息学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689556A (en) * | 2019-09-09 | 2020-01-14 | 苏州臻迪智能科技有限公司 | Tracking method and device and intelligent equipment |
| CN110689559A (en) * | 2019-09-30 | 2020-01-14 | 长安大学 | Visual target tracking method based on dense convolutional network characteristics |
| CN110689559B (en) * | 2019-09-30 | 2022-08-12 | 长安大学 | Visual target tracking method based on dense convolutional network characteristics |
| CN112749602A (en) * | 2019-10-31 | 2021-05-04 | 北京市商汤科技开发有限公司 | Target query method, device, equipment and storage medium |
| CN111091585A (en) * | 2020-03-19 | 2020-05-01 | 腾讯科技（深圳）有限公司 | Target tracking method, device and storage medium |
| CN111091585B (en) * | 2020-03-19 | 2020-07-17 | 腾讯科技（深圳）有限公司 | Target tracking method, device and storage medium |
| CN111415373A (en) * | 2020-03-20 | 2020-07-14 | 北京以萨技术股份有限公司 | Target tracking and segmenting method, system and medium based on twin convolutional network |
| CN111724409A (en) * | 2020-05-18 | 2020-09-29 | 浙江工业大学 | Target tracking method based on densely connected twin neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110188753A (en) | One kind being based on dense connection convolutional neural networks target tracking algorism | |
| CN109949255B (en) | Image reconstruction method and device | |
| CN101739712B (en) | Video-based 3D human face expression cartoon driving method | |
| CN109543754A (en) | The parallel method of target detection and semantic segmentation based on end-to-end deep learning | |
| CN108319972A (en) | A kind of end-to-end difference online learning methods for image, semantic segmentation | |
| CN107563381A (en) | The object detection method of multiple features fusion based on full convolutional network | |
| CN107704877A (en) | A kind of image privacy cognitive method based on deep learning | |
| CN107609525A (en) | Remote Sensing Target detection method based on Pruning strategy structure convolutional neural networks | |
| CN111767847B (en) | Pedestrian multi-target tracking method integrating target detection and association | |
| CN109829934A (en) | A kind of novel image tracking algorithm based on twin convolutional network | |
| CN112967178B (en) | Image conversion method, device, equipment and storage medium | |
| CN110334779A (en) | A kind of multi-focus image fusing method based on PSPNet detail extraction | |
| CN104751172B (en) | The sorting technique of Polarimetric SAR Image based on denoising autocoding | |
| CN110473231A (en) | A kind of method for tracking target of the twin full convolutional network with anticipation formula study more new strategy | |
| CN111695523B (en) | Double-flow convolutional neural network action recognition method based on skeleton space-time and dynamic information | |
| CN104376334B (en) | A kind of pedestrian comparison method of multi-scale feature fusion | |
| CN111814889A (en) | Single-stage target detection method using anchor-frame-free module and enhanced classifier | |
| CN106991380A (en) | A kind of preprocess method based on vena metacarpea image | |
| CN114419413A (en) | Method for constructing sensing field self-adaptive transformer substation insulator defect detection neural network | |
| Meng et al. | Hyperspectral image classification with mixed link networks | |
| CN108268603A (en) | A kind of community discovery method based on core member's identification | |
| CN111274905A (en) | AlexNet and SVM combined satellite remote sensing image land use change detection method | |
| CN113706581A (en) | Target tracking method based on residual channel attention and multilevel classification regression | |
| CN107944437A (en) | A kind of Face detection method based on neutral net and integral image | |
| CN114550308B (en) | Human skeleton action recognition method based on space-time diagram |
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: 20190830 |