CN108171752A - A kind of sea ship video detection and tracking based on deep learning - Google Patents
A kind of sea ship video detection and tracking based on deep learning Download PDFInfo
- Publication number
- CN108171752A CN108171752A CN201711461818.3A CN201711461818A CN108171752A CN 108171752 A CN108171752 A CN 108171752A CN 201711461818 A CN201711461818 A CN 201711461818A CN 108171752 A CN108171752 A CN 108171752A
- Authority
- CN
- China
- Prior art keywords
- layer
- target
- feature
- frame
- convolutional
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
- G06T7/337—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving reference images or patches
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of sea ship video detection and tracking based on deep learning, (1) convolution model feature extractions:It is learned automatically using depth convolutional neural networks and takes target ship feature, each characteristic layer goes extraction feature using a series of 3x3 convolution kernels, and the convolutional layer of shallow-layer mainly extracts minutia, and the convolutional layer of deep layer mainly extracts abstract global contour feature;(2) multiple dimensioned full convolution model target positioning and classification:The big measure feature obtained using trunk convolutional network, is reintegrated in multiple dimensioned convolutional layer, constructs the target signature pyramid network that scale successively decreases, each layer represents different size of target signature, fusion comparison is carried out between characteristic layer;(3) frame target following before and after full connection:These features are cascaded respectively by selecting the feature in target proximity region after convolutional layer extraction feature and input full articulamentum by present frame and previous frame image.The present invention effectively ensures ship detection and the real-time and stability of tracking.
Description
Technical field
The invention belongs to sea ship video detections and tracking technique field more particularly to a kind of sea ship video detection
With the method for tracking.
Background technology
The sea visible ray or the target detection of infrared image obtained based on imaging sensor is had safely in coast defence extensively should
With.Detection zone is limited to offshore or out of, certain marine point visual range, and sea motion carrier and marine marker are equal
Good system deployment place can be used as.
Traditional video image sea-surface target detection method mainly has the Weak Classifier cascade side based on boosting frames
Method and the target matching method based on support vector machines.But the feature representation ability of these methods is limited, can not cope with Sea Surface Ship
It is illumination variation under the conditions of the variation of shape is various, ship occurs in video the variation of angular dimension, different weather, red
Outer image imaging effect poor with respect to visible ray etc..Therefore it is serious to the missing inspection of target and flase drop situation.
On the other hand, monitor video usually requires that high real-time and fine definition, however, these two aspects is shifting
Relationship.The resolution ratio of high clear video image be 1920*1080, such input picture, traditional target based on sliding window
Detection method will generate million grades of couple candidate detection window, take seriously, can not accomplish in real time.
Invention content
It is an object of the invention to solve above-mentioned technical problem, a kind of sea ship video inspection based on deep learning is provided
Survey and tracking, the ship in video image is detected using the method based on deep learning convolutional neural networks with
Track simultaneously carries out high parallel computation using GPU, to ensure real-time and stability.
Technical scheme is as follows:
A kind of sea ship video detection and tracking based on deep learning, carry out according to the following steps:
(1) quickly convolution model feature extraction:Clarification of objective is extracted by trunk convolutional network;Using depth convolution
Neural network is learned and takes target ship feature automatically, each characteristic layer, all goes extraction feature using a series of 3x3 convolution kernels,
Wherein, the convolutional layer of shallow-layer mainly extracts minutia, and the convolutional layer of deep layer mainly extracts abstract global contour feature, and rolls up
Following sample level connection between lamination;
(2) multiple dimensioned full convolution model target positioning and classification:The big measure feature obtained using trunk convolutional network, more
Scale convolutional layer is reintegrated, and constructs the target signature pyramid network that scale successively decreases, and each layer represents different size of mesh
Target feature carries out fusion comparison between these characteristic layers, to determine the size and location of target;
(3) frame target following before and after full connection:After present frame and previous frame image extract feature by convolutional layer respectively,
The feature in target proximity region is selected, these features are cascaded and inputs full articulamentum.
Further, video frame is normalized to the input for being converted into fixed size as Feature Selection Model, soon
The convolution Feature Selection Model of speed uses 18 layers of convolutional neural networks structure, by multiple continuous convolutional layer interval ponds layer group
Into.
Further, the multiple continuous convolutional layer interval pond layer composition structure is:Convolutional layer * 2- ponds layer-volume
Lamination * 2- ponds layer-convolutional layer * 3- ponds layer-convolutional layer * 3- ponds layer-convolutional layer * 3- ponds layer.
Further, convolutional layer uses the convolution kernel of 3*3, with 1 convolution last layer input data of step-length, then plus one
B is biased, linear transformation is converted into nonlinear transformation by correcting linear unit R eLU activation primitives,, wherein, which n represents
Layer network, X are characteristics of image, and W is convolution kernel weight, and g (x) is activation primitive.
Further, the pond layer uses 2*2 convolution, and with 2 convolution last layer input data of step-length, sample mode is most
Input feature vector figure is mapped to a characteristic pattern for reducing 4 times by big value sampling, each pond layer.
Further, pyramidal each layer of the feature multiple dimensioned full convolution model obtained is set close as output layer
Identical candidate frame is spent, the candidate frame on pyramid top corresponds to target frame larger in original image, the candidate frame pair at pyramid bottom
Should in original image smaller target frame.
Further, Multi-task loss calculating is carried out to the candidate frame;Each ROI outputs discrete probability distribution
For:It is with the bounding box displacements returned, wherein, k represents the index of classification, and parameter x, y refers to relative to candidate target
The translational movement of frame Scale invariant, parameter w, h refer in log space relative to the Gao Yukuan of candidate target frame;Loss function represents
For:, wherein, it is true classification, is Classification Loss, is positioning loss, is the candidate frame to match with ground truth box
Number, be weight;Classification Loss is calculated using softmax functions, the real value DUAL PROBLEMS OF VECTOR MAPPINGs that a k is tieed up are into one
The constant vector of range 0-1 is classified according to its size;
Further, in known previous frame object space, region of the present frame near the position where object is original
It scans for, in t-1 frames, it is assumed that target position is (cx, cy), and size is (w, h), then extracting a block size is
The image block of (2w, 2h) is input in CNN;In t frames, equally also centered on (cx, cy), extraction size is (2w, 2h)
Image block, be input in CNN;What full articulamentum learnt is a complicated feature comparison function, exports the opposite of target
Movement, the then output of full articulamentum are connected to the layer of 4 nodes, represent respectively bounding-box top left co-ordinates with
Target length and width, so as to export the position of target.
Beneficial effects of the present invention:Realize imaging sensor to sea ship from motion tracking, detecting and tracking algorithm
Accuracy and robustness are all critically important, for this purpose, the present invention uses the method based on deep learning convolutional neural networks to video figure
Ship as in is detected and tracks, while carries out high parallel computation using GPU, effectively ensures ship detection and tracking
Real-time and stability.
Specific embodiment
Embodiments of the present invention are illustrated by particular specific embodiment below, those skilled in the art can be by this explanation
Content disclosed by book understands other advantages and effect of the present invention easily.
With reference to embodiment, the invention will be further described:
The flow of sea ship video detection and tracking based on deep learning is as follows:
1st, quick convolution model feature extraction:
Clarification of objective is extracted by trunk convolutional network.The spy that traditional target detection tracking method passes through engineer
Sign modeling target, such as LBP, HOG, ICF, the ability to express of feature are limited.This technology invention is using depth convolutional neural networks
Automatic learn takes target ship feature, has merged the characteristic information on multiple abstraction hierarchies.Each characteristic layer all employs a system
The 3x3 convolution kernels of row go extraction feature, and the convolutional layer of shallow-layer mainly extracts minutia, and the convolutional layer of deep layer mainly extracts abstract
Global contour feature.Following sample level connection, makes characteristic layer scale gradually reduce, forms feature pyramid between convolutional layer.
Video frame is normalized to the input for being converted into fixed size as Feature Selection Model.Quick convolution feature extraction mould
Type uses 18 layers of convolutional neural networks structure, is made of multiple continuous convolutional layer interval ponds layer.Appropriate network depth
Guaranteeing to extract the features of enough complexity, calculating speed is quick enough simultaneously.Concrete structure is as follows:Convolutional layer * 2- ponds layer-volume
Lamination * 2- ponds layer-convolutional layer * 3- ponds layer-convolutional layer * 3- ponds layer-convolutional layer * 3- ponds layer.
Convolutional layer uses the convolution kernel of 3*3, with 1 convolution last layer input data of step-length, then plus a biasing b, leads to
It crosses and corrects linear unit R eLU activation primitives linear transformation is converted into nonlinear transformation.Xn=g (WTXn-1+bn), g (x)=
ReLU (x)=max (0, x) (which layer network n represents, and X is characteristics of image, and W is convolution kernel weight, and g (x) is activation primitive).
For linear function, the ability to express of ReLU is stronger, is especially embodied in depth network, and for nonlinear function and
Since the gradient in non-negative section is constant gradient disappearance problem is not present so that the convergence rate of model maintains in speech, ReLU
In a stable state.
Pond layer uses 2*2 convolution, with 2 convolution last layer input data of step-length.Sample mode is sampled for maximum value.Each
Input feature vector figure is mapped to a characteristic pattern for reducing 4 times by pond layer.
2nd, multiple dimensioned full convolution model target positioning and classification:
The big measure feature obtained using trunk convolutional network, is reintegrated in multiple dimensioned convolutional layer, constructs scale and successively decrease
Target signature pyramid network, each layer represents different size of clarification of objective, merged between these characteristic layers
Comparison, the size and location of final determining target.
Pyramidal each layer of the feature that multiple dimensioned full convolution model is obtained is as output layer, the identical time of setting density
Frame is selected, the candidate frame on pyramid top corresponds to target frame larger in original image, and the candidate frame at pyramid bottom corresponds to original image
In smaller target frame.Multi-task loss calculating is carried out to these candidate frames.Discrete probability distribution is exported to each ROI:
P=(p0,...,pk) and bounding box return displacementWherein k represents the index of classification, the first two
Parameter x, y refer to the translation relative to candidate target frame Scale invariant, latter two parameter w, and h refers in log space relative to time
Select the Gao Yukuan of target frame.Loss function is expressed as:Wherein k*It is true
Real classification, LconfIt is Classification Loss, LlocIt is positioning loss, N is with the ground truth box candidate frames to match
Number, α is weight.Classification Loss is calculated using softmax functions, the real value DUAL PROBLEMS OF VECTOR MAPPINGs of k dimension into a model
The constant vector of 0-1 is enclosed, is classified according to its size.
Work as | x |<When 1, smoothL1(x) equal to 0.5x2, when x takes its residual value, smoothL1(x) it is equal to | x | -0.5.
3rd, frame target following before and after full connection:
After present frame and previous frame image extract feature by convolutional layer respectively, the spy in target proximity region is selected
Sign, these features are cascaded and input full articulamentum, the purpose of full articulamentum is to compare the target that previous frame detects
The feature of feature and present frame, with find target be moved to where.Known previous frame object space, present frame is in object original
The region near position where coming scans for.In t-1 frames, it is assumed that target position is (cx, cy), and size is
(w, h) then extracts the image block that a block size is (2w, 2h) and is input in CNN.In t frames, also centered on (cx, cy),
The image block that size is (2w, 2h) is extracted, is input in CNN.What full articulamentum learnt is that a complicated feature compares letter
Number, exports the relative motion of target.Then the output of full articulamentum is connected to the layer of 4 nodes, represents respectively
Bounding-box top left co-ordinates and target length and width, to export the position of target.
The above-described embodiments merely illustrate the principles and effects of the present invention, and is not intended to limit the present invention.It is any ripe
The personage for knowing this technology all can carry out modifications and changes under the spirit and scope without prejudice to the present invention to above-described embodiment.Cause
This, those of ordinary skill in the art is complete without departing from disclosed spirit and institute under technological thought such as
Into all equivalent modifications or change, should by the present invention claim be covered.
Claims (8)
1. a kind of sea ship video detection and tracking based on deep learning, which is characterized in that carry out according to the following steps:
(1) quickly convolution model feature extraction:Clarification of objective is extracted by trunk convolutional network;Using depth convolutional Neural
Network is learned and takes target ship feature automatically, each characteristic layer, all goes extraction feature using a series of 3x3 convolution kernels, wherein,
The convolutional layer of shallow-layer mainly extracts minutia, and the convolutional layer of deep layer mainly extracts abstract global contour feature, and convolutional layer
Between the connection of following sample level;
(2) multiple dimensioned full convolution model target positioning and classification:The big measure feature obtained using trunk convolutional network, multiple dimensioned
Convolutional layer is reintegrated, and constructs the target signature pyramid network that scale successively decreases, and each layer represents different size of target
Feature carries out fusion comparison between these characteristic layers, to determine the size and location of target;
(3) frame target following before and after full connection:After present frame and previous frame image extract feature by convolutional layer respectively, choose
Go out the feature in target proximity region, these features are cascaded and input full articulamentum.
2. sea ship video detection and tracking based on deep learning according to claim 1, which is characterized in that will
The input for being converted into fixed size as Feature Selection Model, quick convolution Feature Selection Model is normalized in video frame
Using 18 layers of convolutional neural networks structure, it is made of multiple continuous convolutional layer interval ponds layer.
3. sea ship video detection and tracking based on deep learning according to claim 2, which is characterized in that institute
Stating multiple continuous convolutional layer interval ponds layer composition structure is:Convolutional layer * 2- ponds layer-convolutional layer * 2- ponds layer-convolution
Layer * 3- ponds layer-convolutional layer * 3- ponds layer-convolutional layer * 3- ponds layer.
4. sea ship video detection and tracking based on deep learning according to claim 3, which is characterized in that volume
Lamination uses the convolution kernel of 3*3, with 1 convolution last layer input data of step-length, then plus a biasing b, by correcting linearly
Linear transformation is converted to nonlinear transformation, X by unit R eLU activation primitivesn=g (WTXn-1+bn), g (x)=ReLU (x)=max
(0, x), wherein, which layer network n represents, and X is characteristics of image, and W is convolution kernel weight, and g (x) is activation primitive.
5. sea ship video detection and tracking based on deep learning according to claim 4, which is characterized in that institute
Pond layer is stated using 2*2 convolution, with 2 convolution last layer input data of step-length, sample mode is sampled for maximum value, each pond layer
Input feature vector figure is mapped to a characteristic pattern for reducing 4 times.
6. sea ship video detection and tracking based on deep learning according to any one of claim 1-5, special
Sign is that pyramidal each layer of the feature for obtaining multiple dimensioned full convolution model is as output layer, the identical time of setting density
Frame is selected, the candidate frame on pyramid top corresponds to target frame larger in original image, and the candidate frame at pyramid bottom corresponds to original image
In smaller target frame.
7. sea ship video detection and tracking based on deep learning according to claim 6, which is characterized in that right
The candidate frame carries out Multi-taskloss calculating, and each ROI outputs discrete probability distribution is:P=(p0,...,pk) and
Boundingbox return displacement beWherein, k represents the index of classification, and parameter x, y refers to relative to time
The translational movement of target frame Scale invariant is selected, parameter w, h refers in log space relative to the Gao Yukuan of candidate target frame;Lose letter
Number is expressed as:Wherein, k*It is true classification, LconfIt is Classification Loss,
LlocIt is positioning loss, N is the number of the candidate frame to match with groundtruthbox, and α is weight;Using softmax functions
Calculate Classification Loss, the realvalue DUAL PROBLEMS OF VECTOR MAPPINGs of k dimension into the constant vector of a range 0-1, according to its size come
Classify;
8. the sea ship video detection and tracking, feature according to claim 1 or 7 based on deep learning exist
In in known previous frame object space, region of the present frame near the position where object is original scans for, in t-
In 1 frame, it is assumed that target position is (cx, cy), and size is (w, h), then extracts the image block that a block size is (2w, 2h)
It is input in CNN;In t frames, equally also centered on (cx, cy), extraction size is the image block of (2w, 2h), is input to
In CNN;What full articulamentum learnt is a complicated feature comparison function, exports the relative motion of target, then full connection
The output of layer is connected to the layer of 4 nodes, represents bounding-box top left co-ordinates and target length and width respectively, so as to defeated
Go out the position of target.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711461818.3A CN108171752A (en) | 2017-12-28 | 2017-12-28 | A kind of sea ship video detection and tracking based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711461818.3A CN108171752A (en) | 2017-12-28 | 2017-12-28 | A kind of sea ship video detection and tracking based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108171752A true CN108171752A (en) | 2018-06-15 |
Family
ID=62519433
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711461818.3A Pending CN108171752A (en) | 2017-12-28 | 2017-12-28 | A kind of sea ship video detection and tracking based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108171752A (en) |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108831161A (en) * | 2018-06-27 | 2018-11-16 | 深圳大学 | A kind of traffic flow monitoring method, intelligence system and data set based on unmanned plane |
CN109190442A (en) * | 2018-06-26 | 2019-01-11 | 杭州雄迈集成电路技术有限公司 | A kind of fast face detecting method based on depth cascade convolutional neural networks |
CN109242019A (en) * | 2018-09-01 | 2019-01-18 | 哈尔滨工程大学 | A kind of water surface optics Small object quickly detects and tracking |
CN109345559A (en) * | 2018-08-30 | 2019-02-15 | 西安电子科技大学 | Expand the motion target tracking method with depth sorting network based on sample |
CN109446978A (en) * | 2018-10-25 | 2019-03-08 | 哈尔滨工程大学 | Based on the winged maneuvering target tracking method for staring satellite complex scene |
CN109472298A (en) * | 2018-10-19 | 2019-03-15 | 天津大学 | Depth binary feature pyramid for the detection of small scaled target enhances network |
CN109711427A (en) * | 2018-11-19 | 2019-05-03 | 深圳市华尊科技股份有限公司 | Object detection method and Related product |
CN109737974A (en) * | 2018-12-14 | 2019-05-10 | 中国科学院深圳先进技术研究院 | A kind of 3D navigational semantic map updating method, device and equipment |
CN109740552A (en) * | 2019-01-09 | 2019-05-10 | 上海大学 | A kind of method for tracking target based on Parallel Signature pyramid neural network |
CN109740665A (en) * | 2018-12-29 | 2019-05-10 | 珠海大横琴科技发展有限公司 | Shielded image ship object detection method and system based on expertise constraint |
CN109784278A (en) * | 2019-01-17 | 2019-05-21 | 上海海事大学 | The small and weak moving ship real-time detection method in sea based on deep learning |
CN109800716A (en) * | 2019-01-22 | 2019-05-24 | 华中科技大学 | One kind being based on the pyramidal Oceanic remote sensing image ship detecting method of feature |
CN109840921A (en) * | 2019-01-29 | 2019-06-04 | 北京三快在线科技有限公司 | The determination method, apparatus and unmanned equipment of unmanned task result |
CN109859163A (en) * | 2018-12-19 | 2019-06-07 | 重庆邮电大学 | A kind of LCD defect inspection method based on feature pyramid convolutional neural networks |
CN110021033A (en) * | 2019-02-22 | 2019-07-16 | 广西师范大学 | A kind of method for tracking target based on the twin network of pyramid |
CN110060508A (en) * | 2019-04-08 | 2019-07-26 | 武汉理工大学 | A kind of ship automatic testing method for inland river bridge zone |
CN110097028A (en) * | 2019-05-14 | 2019-08-06 | 河北工业大学 | Crowd's accident detection method of network is generated based on three-dimensional pyramid diagram picture |
CN110675377A (en) * | 2019-09-20 | 2020-01-10 | 国网湖北省电力有限公司电力科学研究院 | State monitoring system and method for substation relay protection device |
CN110689081A (en) * | 2019-09-30 | 2020-01-14 | 中国科学院大学 | Weak supervision target classification and positioning method based on bifurcation learning |
CN110874953A (en) * | 2018-08-29 | 2020-03-10 | 杭州海康威视数字技术股份有限公司 | Area alarm method and device, electronic equipment and readable storage medium |
CN110956108A (en) * | 2019-11-22 | 2020-04-03 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
CN111368690A (en) * | 2020-02-28 | 2020-07-03 | 珠海大横琴科技发展有限公司 | Deep learning-based video image ship detection method and system under influence of sea waves |
CN111652903A (en) * | 2020-05-22 | 2020-09-11 | 重庆理工大学 | Pedestrian target tracking method based on convolution correlation network in automatic driving scene |
CN112085767A (en) * | 2020-08-28 | 2020-12-15 | 安徽清新互联信息科技有限公司 | Passenger flow statistical method and system based on deep optical flow tracking |
CN112465811A (en) * | 2020-12-16 | 2021-03-09 | 国网四川省电力公司电力科学研究院 | High-precision dynamic target positioning method and device for electric power operation site |
CN113298080A (en) * | 2021-07-26 | 2021-08-24 | 城云科技(中国)有限公司 | Target detection enhancement model, target detection method, target detection device and electronic device |
CN113375568A (en) * | 2021-05-12 | 2021-09-10 | 苏州阿普奇物联网科技有限公司 | Metal wiredrawing polishing defect detection method based on laser scanning |
CN114241003A (en) * | 2021-12-14 | 2022-03-25 | 成都阿普奇科技股份有限公司 | All-weather lightweight high-real-time sea surface ship detection and tracking method |
CN114581796A (en) * | 2022-01-19 | 2022-06-03 | 上海土蜂科技有限公司 | Target tracking system, method and computer device thereof |
CN114820716A (en) * | 2022-06-29 | 2022-07-29 | 中兴软件技术(南昌)有限公司 | Target tracking method and system based on computer vision |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160140424A1 (en) * | 2014-11-13 | 2016-05-19 | Nec Laboratories America, Inc. | Object-centric Fine-grained Image Classification |
CN106067022A (en) * | 2016-05-28 | 2016-11-02 | 北方工业大学 | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm |
CN106886755A (en) * | 2017-01-19 | 2017-06-23 | 北京航空航天大学 | A kind of intersection vehicles system for detecting regulation violation based on Traffic Sign Recognition |
CN106980895A (en) * | 2017-02-22 | 2017-07-25 | 中国科学院自动化研究所 | Convolutional neural networks Forecasting Methodology based on rotary area |
CN107103279A (en) * | 2017-03-09 | 2017-08-29 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of passenger flow counting method under vertical angle of view based on deep learning |
CN107274433A (en) * | 2017-06-21 | 2017-10-20 | 吉林大学 | Method for tracking target, device and storage medium based on deep learning |
CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
-
2017
- 2017-12-28 CN CN201711461818.3A patent/CN108171752A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160140424A1 (en) * | 2014-11-13 | 2016-05-19 | Nec Laboratories America, Inc. | Object-centric Fine-grained Image Classification |
CN106067022A (en) * | 2016-05-28 | 2016-11-02 | 北方工业大学 | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm |
CN106886755A (en) * | 2017-01-19 | 2017-06-23 | 北京航空航天大学 | A kind of intersection vehicles system for detecting regulation violation based on Traffic Sign Recognition |
CN106980895A (en) * | 2017-02-22 | 2017-07-25 | 中国科学院自动化研究所 | Convolutional neural networks Forecasting Methodology based on rotary area |
CN107103279A (en) * | 2017-03-09 | 2017-08-29 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of passenger flow counting method under vertical angle of view based on deep learning |
CN107274433A (en) * | 2017-06-21 | 2017-10-20 | 吉林大学 | Method for tracking target, device and storage medium based on deep learning |
CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
Non-Patent Citations (4)
Title |
---|
WEI LIU 等: "SSD: Single Shot MultiBox Detector", 《COMPUTER VISION–ECCV 2016》 * |
YUANYUAN WANG 等: "Combing Single Shot Multibox Detector with transfer learning for ship detection using Chinese Gaofen-3 images", 《2017 PROGRESS IN ELECTROMAGNETICS RESEARCH SYMPOSIUM - FALL》 * |
YUANYUAN WANG 等: "Combining single shot multibox detector with transfer learning for ship detection using Sentinel-1 images", 《2017 SAR IN BIG DATA ERA: MODELS, METHODS AND APPLICATIONS》 * |
浩瀚之水_CSDN: "CNN目标检测(三): SSD详解", 《HTTPS://BLOG.CSDN.NET/A8039974/ARTICLE/DETAILS/77592395》 * |
Cited By (43)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190442A (en) * | 2018-06-26 | 2019-01-11 | 杭州雄迈集成电路技术有限公司 | A kind of fast face detecting method based on depth cascade convolutional neural networks |
CN109190442B (en) * | 2018-06-26 | 2021-07-06 | 杭州雄迈集成电路技术股份有限公司 | Rapid face detection method based on deep cascade convolution neural network |
CN108831161A (en) * | 2018-06-27 | 2018-11-16 | 深圳大学 | A kind of traffic flow monitoring method, intelligence system and data set based on unmanned plane |
CN110874953A (en) * | 2018-08-29 | 2020-03-10 | 杭州海康威视数字技术股份有限公司 | Area alarm method and device, electronic equipment and readable storage medium |
CN109345559A (en) * | 2018-08-30 | 2019-02-15 | 西安电子科技大学 | Expand the motion target tracking method with depth sorting network based on sample |
CN109345559B (en) * | 2018-08-30 | 2021-08-06 | 西安电子科技大学 | Moving target tracking method based on sample expansion and depth classification network |
CN109242019A (en) * | 2018-09-01 | 2019-01-18 | 哈尔滨工程大学 | A kind of water surface optics Small object quickly detects and tracking |
CN109242019B (en) * | 2018-09-01 | 2022-05-17 | 哈尔滨工程大学 | Rapid detection and tracking method for optical small target on water surface |
CN109472298A (en) * | 2018-10-19 | 2019-03-15 | 天津大学 | Depth binary feature pyramid for the detection of small scaled target enhances network |
CN109472298B (en) * | 2018-10-19 | 2021-06-01 | 天津大学 | Deep bidirectional feature pyramid enhanced network for small-scale target detection |
CN109446978A (en) * | 2018-10-25 | 2019-03-08 | 哈尔滨工程大学 | Based on the winged maneuvering target tracking method for staring satellite complex scene |
CN109711427A (en) * | 2018-11-19 | 2019-05-03 | 深圳市华尊科技股份有限公司 | Object detection method and Related product |
CN109737974A (en) * | 2018-12-14 | 2019-05-10 | 中国科学院深圳先进技术研究院 | A kind of 3D navigational semantic map updating method, device and equipment |
CN109859163A (en) * | 2018-12-19 | 2019-06-07 | 重庆邮电大学 | A kind of LCD defect inspection method based on feature pyramid convolutional neural networks |
CN109740665A (en) * | 2018-12-29 | 2019-05-10 | 珠海大横琴科技发展有限公司 | Shielded image ship object detection method and system based on expertise constraint |
CN109740552A (en) * | 2019-01-09 | 2019-05-10 | 上海大学 | A kind of method for tracking target based on Parallel Signature pyramid neural network |
CN109784278A (en) * | 2019-01-17 | 2019-05-21 | 上海海事大学 | The small and weak moving ship real-time detection method in sea based on deep learning |
CN109784278B (en) * | 2019-01-17 | 2020-09-29 | 上海海事大学 | Deep learning-based marine small and weak motion ship real-time detection method |
CN109800716A (en) * | 2019-01-22 | 2019-05-24 | 华中科技大学 | One kind being based on the pyramidal Oceanic remote sensing image ship detecting method of feature |
CN109840921A (en) * | 2019-01-29 | 2019-06-04 | 北京三快在线科技有限公司 | The determination method, apparatus and unmanned equipment of unmanned task result |
CN110021033B (en) * | 2019-02-22 | 2023-05-30 | 新疆交投科技有限责任公司 | Target tracking method based on pyramid twin network |
CN110021033A (en) * | 2019-02-22 | 2019-07-16 | 广西师范大学 | A kind of method for tracking target based on the twin network of pyramid |
CN110060508A (en) * | 2019-04-08 | 2019-07-26 | 武汉理工大学 | A kind of ship automatic testing method for inland river bridge zone |
CN110097028B (en) * | 2019-05-14 | 2021-03-30 | 河北工业大学 | Crowd abnormal event detection method based on three-dimensional pyramid image generation network |
CN110097028A (en) * | 2019-05-14 | 2019-08-06 | 河北工业大学 | Crowd's accident detection method of network is generated based on three-dimensional pyramid diagram picture |
CN110675377A (en) * | 2019-09-20 | 2020-01-10 | 国网湖北省电力有限公司电力科学研究院 | State monitoring system and method for substation relay protection device |
CN110675377B (en) * | 2019-09-20 | 2022-03-22 | 国网湖北省电力有限公司电力科学研究院 | State monitoring system and method for substation relay protection device |
CN110689081A (en) * | 2019-09-30 | 2020-01-14 | 中国科学院大学 | Weak supervision target classification and positioning method based on bifurcation learning |
CN110956108A (en) * | 2019-11-22 | 2020-04-03 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
CN110956108B (en) * | 2019-11-22 | 2023-04-18 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
CN111368690B (en) * | 2020-02-28 | 2021-03-02 | 珠海大横琴科技发展有限公司 | Deep learning-based video image ship detection method and system under influence of sea waves |
CN111368690A (en) * | 2020-02-28 | 2020-07-03 | 珠海大横琴科技发展有限公司 | Deep learning-based video image ship detection method and system under influence of sea waves |
CN111652903A (en) * | 2020-05-22 | 2020-09-11 | 重庆理工大学 | Pedestrian target tracking method based on convolution correlation network in automatic driving scene |
CN111652903B (en) * | 2020-05-22 | 2023-09-08 | 重庆理工大学 | Pedestrian target tracking method based on convolution association network in automatic driving scene |
CN112085767A (en) * | 2020-08-28 | 2020-12-15 | 安徽清新互联信息科技有限公司 | Passenger flow statistical method and system based on deep optical flow tracking |
CN112465811A (en) * | 2020-12-16 | 2021-03-09 | 国网四川省电力公司电力科学研究院 | High-precision dynamic target positioning method and device for electric power operation site |
CN113375568A (en) * | 2021-05-12 | 2021-09-10 | 苏州阿普奇物联网科技有限公司 | Metal wiredrawing polishing defect detection method based on laser scanning |
CN113298080A (en) * | 2021-07-26 | 2021-08-24 | 城云科技(中国)有限公司 | Target detection enhancement model, target detection method, target detection device and electronic device |
CN114241003A (en) * | 2021-12-14 | 2022-03-25 | 成都阿普奇科技股份有限公司 | All-weather lightweight high-real-time sea surface ship detection and tracking method |
CN114581796A (en) * | 2022-01-19 | 2022-06-03 | 上海土蜂科技有限公司 | Target tracking system, method and computer device thereof |
CN114581796B (en) * | 2022-01-19 | 2024-04-02 | 上海土蜂科技有限公司 | Target tracking system, method and computer device thereof |
CN114820716A (en) * | 2022-06-29 | 2022-07-29 | 中兴软件技术(南昌)有限公司 | Target tracking method and system based on computer vision |
CN114820716B (en) * | 2022-06-29 | 2022-09-13 | 中兴软件技术(南昌)有限公司 | Target tracking method and system based on computer vision |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108171752A (en) | A kind of sea ship video detection and tracking based on deep learning | |
CN110163187B (en) | F-RCNN-based remote traffic sign detection and identification method | |
CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
CN109711295B (en) | Optical remote sensing image offshore ship detection method | |
CN109241913A (en) | In conjunction with the ship detection method and system of conspicuousness detection and deep learning | |
CN110309808B (en) | Self-adaptive smoke root node detection method in large-scale space | |
CN104573731A (en) | Rapid target detection method based on convolutional neural network | |
CN109657541A (en) | A kind of ship detecting method in unmanned plane image based on deep learning | |
CN105022990A (en) | Water surface target rapid-detection method based on unmanned vessel application | |
CN102298781A (en) | Motion shadow detection method based on color and gradient characteristics | |
CN117253154B (en) | Container weak and small serial number target detection and identification method based on deep learning | |
CN109784278A (en) | The small and weak moving ship real-time detection method in sea based on deep learning | |
CN113255837A (en) | Improved CenterNet network-based target detection method in industrial environment | |
CN113435319B (en) | Classification method combining multi-target tracking and pedestrian angle recognition | |
CN110334656A (en) | Multi-source Remote Sensing Images Clean water withdraw method and device based on information source probability weight | |
CN105405138A (en) | Water surface target tracking method based on saliency detection | |
CN113743322A (en) | Offshore ship detection method based on improved YOLOv3 algorithm | |
CN109635726A (en) | A kind of landslide identification method based on the symmetrical multiple dimensioned pond of depth network integration | |
CN114332921A (en) | Pedestrian detection method based on improved clustering algorithm for Faster R-CNN network | |
Wang et al. | Human detection based on improved mask R-CNN | |
Zhao et al. | Boundary regularized building footprint extraction from satellite images using deep neural network | |
Liu et al. | [Retracted] Self‐Correction Ship Tracking and Counting with Variable Time Window Based on YOLOv3 | |
Chen et al. | Research on fast recognition method of complex sorting images based on deep learning | |
CN107230210A (en) | A kind of fast partition method of remote sensing images harbour waterborne target | |
Gan et al. | Object detection in remote sensing images with mask R-CNN |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |