CN109697727A - Method for tracking target, system and storage medium based on correlation filtering and metric learning - Google Patents
Method for tracking target, system and storage medium based on correlation filtering and metric learning Download PDFInfo
- Publication number
- CN109697727A CN109697727A CN201811425898.1A CN201811425898A CN109697727A CN 109697727 A CN109697727 A CN 109697727A CN 201811425898 A CN201811425898 A CN 201811425898A CN 109697727 A CN109697727 A CN 109697727A
- Authority
- CN
- China
- Prior art keywords
- target
- tracking
- correlation filtering
- metric learning
- confidence level
- 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
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of method for tracking target based on correlation filtering and metric learning, system and storage medium, which includes: step S1: obtaining the size and location of target in the first frame of tracking video；Step S2: the template of target signature and negative sample similar with target and training correlation filtering is obtained according to obtained target information；Step S3: by the position for the template prediction target that training obtains, and the confidence level that figure is this time predicted according to response is to determine whether be accurately positioned target using metric learning network.The beneficial effects of the present invention are: the present invention is directed to the characteristics of target tracking video, first present frame is predicted using correlation filtering model, is found out and candidate region similar in target；For these candidate regions, the similarity of itself and target is judged using depth network；For target tracking, it is the optimization to correlation filtering model, joined metric learning algorithm, propose using metric learning and optimize the method for correlation filtering result；So as to accurately and real-time track target.
Description
Technical field
The present invention relates to real-time target following technical fields, more particularly to one kind to be based on correlation filtering in the video sequence
With metric learning technology to the tracking of target, system and storage medium.
Background technique
It is one of important research direction of computer vision field that vision, which tracks (Visual Tracking) technology,.Vision
Tracking extracts the feature in image using the relevant knowledges such as machine learning, image procossing, deep learning, then to feature modeling,
It is finally trained and judges target position to each frame in the video sequence using model, realize the tracking to target.Target
Tracking also has many application scenarios, such as vehicle and pedestrian's monitoring, intelligent security guard.Target tracking is mostly applied in video monitoring
In.It needs the flow to vehicle and pedestrian to be monitored in smart city, needs in monitoring to vehicle and pedestrian from appearance
This section of process to disappearance is tracked.
Visual Tracking is more and more paid close attention to by people.One typical scene of vision tracking is to track one not
The target object known, it is specified by the target frame (Bounding Box) in first frame.Target following is in the field that has a wide range of applications
Scape, however due to target appearance variation and the variation of ambient enviroment and the presence of similar background disturbing factor, so that model exists
Target is tracked under certain scenes certain difficulty.In recent years, domestic and international researcher continuously improved algorithm to improve tracker
Accuracy and robustness.Target visual tracking is broadly divided into two major classes: generating (generative) model method and differentiation
(discriminative) model method, popular at present is to differentiate class method, is also detecting and tracking (tracking-by-
detection).The method that mainstream algorithm under discrimination model frame has correlation filtering, deep learning.Target tracking algorism research
Difficult point and challenge: the similar interference of application environment, the background of actual complex, the variation of illumination condition, block etc. extraneous factors and
Target carriage change, appearance deformation, dimensional variation, plane internal rotation, go out the visual field, quickly movement and movement mould at plane external rotation
Paste etc..And when target tracking algorism puts into practical application, an inevitable problem --- real time problems also right and wrong
Normal is important.Exactly there are these problems, just algorithm research is made to be filled with difficult point and challenge.
Summary of the invention
The present invention provides a kind of method for tracking target based on correlation filtering and metric learning, including successively execute as follows
Step:
Step S1: the size and location of target is obtained in the first frame of tracking video；
Step S2: target signature and negative sample similar to target are obtained according to obtained target information and train related filter
The template of wave；
Step S3: the position of template prediction target obtained by training, and setting of this time being predicted of figure according to response
Reliability；
Step S4: judging the height of confidence level, if confidence level is high, then step S6 is executed, if confidence level is low, then executing
Step S5；
Step S5: the candidate region of different length-width ratios and size is generated in obtained predicted position, and passes through depth network
Accurately to search the position of target；
Step S6: judging whether video sequence is last frame, if so, so terminate, it is no to then follow the steps S2.
As a further improvement of the present invention, the step S2 includes the following steps:
Step S21: target signature is extracted；
Step S22: negative sample similar with target is found around target；
Step S23: with the template of target signature and similar sample training correlation filtering.
As a further improvement of the present invention, in the step S21, extracting target signature mode is color characteristic or side
Edge feature extraction.
As a further improvement of the present invention, in the step S22, negative sample is determined by the peak value in response diagram
Position.
As a further improvement of the present invention, in the step S4, in response diagram, according to the number and main peak of peak value
Gradient come judge this tracking result confidence level height.
As a further improvement of the present invention, the step S5 includes the following steps:
Step S51: candidate region is generated at random in correlation filtering response setting range to adapt to the length and width of deformation target
Than；
Step S52: the exact position that target is found in judgement is carried out to the candidate region of generation.
The present invention also provides a kind of Target Tracking System based on correlation filtering and metric learning, comprising: memory, place
Reason device and the computer program being stored on the memory, when the computer program is configured to be called by the processor
The step of realizing method of the present invention.
The present invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage has calculating
The step of machine program, the computer program realizes method of the present invention when being configured to be called by processor.
The beneficial effects of the present invention are: the present invention is directed to the characteristics of target tracking video, first come using correlation filtering model
Present frame is predicted, is found out and candidate region similar in target；For these candidate regions, judged using depth network
The similarity of itself and target；For target tracking, it is the optimization to correlation filtering model, joined metric learning algorithm,
The method for proposing correlation filtering and metric learning parallel optimization；So as to accurately and real-time track target.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Fig. 2 is depth measure learning network design drawing of the invention.
Specific embodiment
As shown in Figure 1, the invention discloses a kind of method for tracking target based on correlation filtering and metric learning, including according to
Secondary execution following steps:
Step S1: the size and location of target is obtained in the first frame of tracking video；
Step S2: target signature and negative sample similar to target are obtained according to obtained target information and train related filter
The template of wave；
Step S3: the position of template prediction target obtained by training, and setting of this time being predicted of figure according to response
Reliability；
Step S4: judging the height of confidence level, if confidence level is high, then step S6 is executed, if confidence level is low, then executing
Step S5；
Step S5: the candidate region of different length-width ratios and size is generated in obtained predicted position, and passes through depth network
Accurately to search the position of target；
Step S6: judging whether video sequence is last frame, if so, so terminate, it is no to then follow the steps S2.
The step S2 includes the following steps:
Step S21: target signature is extracted；
Step S22: negative sample similar with target is found around target；
Step S23: with the template of target signature and similar sample training correlation filtering.
In the step S21, extracting target signature mode is color characteristic or Edge Gradient Feature.
In the step S22, the position of negative sample is determined by the peak value in response diagram.
In the step S4, in response diagram, this secondary tracking knot is judged according to the number of peak value and the gradient of main peak
The confidence level height of fruit.
The step S5 includes the following steps:
Step S51: candidate region is generated at random in correlation filtering response setting range to adapt to the length and width of deformation target
Than；
Step S52: the exact position that target is found in judgement is carried out to the candidate region of generation.
In the present invention, correlation filtering class method is one matrix of training to indicate template, passes through trained matrix matching
Target, algorithm include three parts: finding training sample and be added into the training process of correlation filtering template, pass through correlation filtering
Peak value of response judges whether to determine target position using metric learning network, using metric learning.
Find similar with target training sample, in target following we can mark target area in advance, then
The higher region of some scores is found in the result that algorithm obtains again, there is also having for the target area coincidence with label in these regions
It is not overlapped, then registration (IOU) is higher, regards as being positive sample, it is below, it is negative sample.Then, perhaps this can
There is the quantity that a problem is exactly positive sample and be far smaller than negative sample, the effect for training the classifier come so always has
Limit, it may appear that many vacation positive samples, these higher false positive samples of wherein score as so-called difficult example sample, since digging
These difficult example samples have been excavated, it is just that training in these templates is primary, to reinforce the ability that classifier differentiates false positive, in this way
Obtained objective result more meets ours it is contemplated that obtained result is also more fitted with target area.Correlation filtering method method
Can obtain a response diagram, there is the response much put to be not much different in response diagram, but choose result when not into
Row judgement, only only selects that highest position of score value, is likely to result in result erroneous judgement in this way.It, can for this problem
With one metric learning network of training, judged using the similar position of those score values more than this network, passes through judgement front and back
The semantic similarity of target completes the tracking to target between frame, can be improved model accuracy rate.According to current present Research and
Analysis determines the extraction that target signature is carried out by the way of deep learning, and depth characteristic vector more preferable can must express target.
The task of target tracking is to be concatenated the target position in frame each in video.Metric function is for evaluating two
Relationship between sample, how the performance of selected metric function and depth measure network has close relationship.There are many similar
Degree function or distance function may be used as judging the similitude of two vectors.The feature representation of convolutional neural networks study is wide
Their validity is shown in general visual identity task.Convolutional neural networks are used herein to find a public spy
Levy space.The deep structure of neural network can extract more abstract and constant data characteristics, to have than traditional classifier
Higher nicety of grading.Since CNN can effectively have found the space structure between the adjacent block of input data, although not having
The result for having and directly modeling to neighbor dependency, but obtain usually seems smoother.
In the present invention, we want to extract tracking clarification of objective using convolutional neural networks, according to the spy of before and after frames
Sign compares to find the position of target.The network architecture used is relatively shallow, primarily to quickly training and deduction.
When pre-training network one on extensive track file training obtain, this track file includes a large amount of
There is the video of mark, so that the feature acquired is well suited for tracking target.Using every frame training during model training
Method furthers the similarity of similar target.Finally target position is accurately positioned using the pre-training network.
Input picture re-scaling is 64 × 64 by Fig. 2, and network inputs are RGB triple channel.Length is being extracted by dense layer 8
Before 64 global characteristics vector, the size of characteristic pattern is reduced to 1/the 16 of original image by a series of convolution operations.Finally
Target signature is normalized after Dense layers to unit hyper-sphere.The network includes multiple residual blocks.The size of all convolution is
3 × 3, maximum pond is replaced by step-length by 2 convolutional layer.When the spatial resolution of characteristic pattern reduces, then correspondingly increase channel
Quantity.Dropout and batch normalization are used as the means of regularization.Index linear unit is used as the activation letter in all layers
Number.The network proposes that 32 bounding boxes take about 5ms at Nvidia GeForce GTX 1080GPU, it is clear that
It can satisfy requirement of real-time.
To sum up, for target tracking video the characteristics of, the present invention first carry out present frame using correlation filtering model pre-
It surveys, finds out and candidate region similar in target.For these candidate regions, judge that it is similar to target using depth network
Degree.For target tracking, it is the optimization to correlation filtering model, joined metric learning algorithm, propose correlation filtering
With the method for metric learning parallel optimization.It can accurately and real-time track target.
The targeted species for not limiting tracking when tracking specify the target of tracking in the first frame of video sequence, then rear
The tracking apparent to target is completed by designing a model in continuous frame.It is either based on conventional method or deep learning method,
Be required to establish target on model, in conventional methods where using manual feature, ability to express is not strong, but can quickly calculate from
And it can be with on-line training model.Because the target of tracking is not limit type, deep learning is difficult off-line training to fit
Type should largely be tracked.So how not only to have utilized the on-line training of conventional method but also off-line training model in deep learning be added
It is main contents of the invention.The present invention is based on correlation filtering model training part by optimization, joined similar with target
Sample is as negative sample, and depth measure learning network is added to measure the similarity Lifting Modules of the previous frame target of present frame sum
Type accuracy rate.
Negative sample is added in the present invention in correlation filtering.Correlation filtering can be with online updating model, current
Frame trains a template, then with this template in the position of next frame detection target.It can be obtained when adding and surveying target position
To a response diagram.Model may float in the higher negative sample of response during tracking.In this regard, we will respond
It is responded in figure during higher negative sample is added to template training.When on-line training model, select and surrounding mesh
Similar sample is marked as negative sample to increase model to the discriminating power of similar sample.Depth degree is added on the basis of above-mentioned
Measure learning network.It is still likely to the target of plane external rotation or deformation during tracking with losing.In this regard, we are offline
One depth measure network of training tracks the semantic information of target to extract.For example tracking target is behaved, people is standing and is squatting
When target template it is entirely different, but by deep learning, we can be judged and then be passed through to it by semanteme
The similarity of target between before and after frames is judged to complete the tracking to target.
The invention also discloses a kind of Target Tracking System based on correlation filtering and metric learning, comprising: memory, place
Reason device and the computer program being stored on the memory, when the computer program is configured to be called by the processor
The step of realizing method of the present invention.
The invention also discloses a kind of computer readable storage medium, the computer-readable recording medium storage has calculating
The step of machine program, the computer program realizes method of the present invention when being configured to be called by processor.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention
Protection scope.
Claims (8)
1. a kind of method for tracking target based on correlation filtering and metric learning, which is characterized in that including successively executing following step
It is rapid:
Step S1: the size and location of target is obtained in the first frame of tracking video；
Step S2: target signature and negative sample similar with target are obtained according to obtained target information and train correlation filtering
Template；
Step S3: the position by training obtained template prediction target, and the confidence level that figure is this time predicted according to response；
Step S4: judging the height of confidence level, if confidence level is high, then step S6 is executed, if confidence level is low, then executing step
S5；
Step S5: the candidate region of different length-width ratios and size is generated in obtained predicted position, and by depth network come smart
Really search the position of target；
Step S6: judging whether video sequence is last frame, if so, so terminate, it is no to then follow the steps S2.
2. method for tracking target according to claim 1, which is characterized in that the step S2 includes the following steps:
Step S21: target signature is extracted；
Step S22: negative sample similar with target is found around target；
Step S23: with the template of target signature and similar sample training correlation filtering.
3. method for tracking target according to claim 2, which is characterized in that in the step S21, extract target signature
Mode is color characteristic or Edge Gradient Feature.
4. method for tracking target according to claim 2, which is characterized in that in the step S22, by response diagram
Peak value determine the position of negative sample.
5. method for tracking target according to claim 1, which is characterized in that in the step S4, in response diagram, root
According to peak value number and main peak gradient come judge this tracking result confidence level height.
6. method for tracking target according to claim 1, which is characterized in that the step S5 includes the following steps:
Step S51: candidate region is generated at random in correlation filtering response setting range to adapt to the length-width ratio of deformation target；
Step S52: the exact position that target is found in judgement is carried out to the candidate region of generation.
7. a kind of Target Tracking System based on correlation filtering and metric learning characterized by comprising memory, processor
And it is stored in the computer program on the memory, the realization when computer program is configured to be called by the processor
The step of method of any of claims 1-6.
8. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey
The step of sequence, the computer program realizes method of any of claims 1-6 when being configured to be called by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811425898.1A CN109697727A (en) | 2018-11-27 | 2018-11-27 | Method for tracking target, system and storage medium based on correlation filtering and metric learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811425898.1A CN109697727A (en) | 2018-11-27 | 2018-11-27 | Method for tracking target, system and storage medium based on correlation filtering and metric learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109697727A true CN109697727A (en) | 2019-04-30 |
Family
ID=66230201
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811425898.1A Pending CN109697727A (en) | 2018-11-27 | 2018-11-27 | Method for tracking target, system and storage medium based on correlation filtering and metric learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109697727A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197126A (en) * | 2019-05-06 | 2019-09-03 | 深圳岚锋创视网络科技有限公司 | A kind of target tracking method, device and portable terminal |
| CN111242981A (en) * | 2020-01-21 | 2020-06-05 | 北京捷通华声科技股份有限公司 | Tracking method and device for fixed object and security equipment |
| CN111684491A (en) * | 2019-05-31 | 2020-09-18 | 深圳市大疆创新科技有限公司 | Target tracking method, target tracking device and unmanned aerial vehicle |
| US20210203900A1 (en) * | 2020-04-17 | 2021-07-01 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067410A (en) * | 2016-05-27 | 2017-08-18 | 北京理工大学 | A kind of manifold regularization correlation filtering method for tracking target based on augmented sample |
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107424171A (en) * | 2017-07-21 | 2017-12-01 | 华中科技大学 | A kind of anti-shelter target tracking based on piecemeal |
| CN108121945A (en) * | 2017-11-14 | 2018-06-05 | 深圳市深网视界科技有限公司 | A kind of multi-target detection tracking, electronic equipment and storage medium |
| CN108346159A (en) * | 2018-01-28 | 2018-07-31 | 北京工业大学 | A kind of visual target tracking method based on tracking-study-detection |
| CN108717522A (en) * | 2018-04-18 | 2018-10-30 | 上海交通大学 | A kind of human body target tracking method based on deep learning and correlation filtering |
-
2018
- 2018-11-27 CN CN201811425898.1A patent/CN109697727A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067410A (en) * | 2016-05-27 | 2017-08-18 | 北京理工大学 | A kind of manifold regularization correlation filtering method for tracking target based on augmented sample |
| CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
| CN107424171A (en) * | 2017-07-21 | 2017-12-01 | 华中科技大学 | A kind of anti-shelter target tracking based on piecemeal |
| CN108121945A (en) * | 2017-11-14 | 2018-06-05 | 深圳市深网视界科技有限公司 | A kind of multi-target detection tracking, electronic equipment and storage medium |
| CN108346159A (en) * | 2018-01-28 | 2018-07-31 | 北京工业大学 | A kind of visual target tracking method based on tracking-study-detection |
| CN108717522A (en) * | 2018-04-18 | 2018-10-30 | 上海交通大学 | A kind of human body target tracking method based on deep learning and correlation filtering |
Non-Patent Citations (2)
| Title |
|---|
| CHAO MA等: "Hierarchical Convolutional Features for Visual Tracking", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| MARTIN DANELLJAN等: "Convolutional Features for Correlation Filter Based Visual Tracking", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOP (ICCVW)》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197126A (en) * | 2019-05-06 | 2019-09-03 | 深圳岚锋创视网络科技有限公司 | A kind of target tracking method, device and portable terminal |
| CN111684491A (en) * | 2019-05-31 | 2020-09-18 | 深圳市大疆创新科技有限公司 | Target tracking method, target tracking device and unmanned aerial vehicle |
| WO2020237674A1 (en) * | 2019-05-31 | 2020-12-03 | 深圳市大疆创新科技有限公司 | Target tracking method and apparatus, and unmanned aerial vehicle |
| CN111242981A (en) * | 2020-01-21 | 2020-06-05 | 北京捷通华声科技股份有限公司 | Tracking method and device for fixed object and security equipment |
| US20210203900A1 (en) * | 2020-04-17 | 2021-07-01 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
| US11930307B2 (en) * | 2020-04-17 | 2024-03-12 | Beijing Baidu Netcom Science Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Traffic light recognition for complex scene with fusion detections | |
| CN109697727A (en) | Method for tracking target, system and storage medium based on correlation filtering and metric learning | |
| CN110135296A (en) | Airfield runway FOD detection method based on convolutional neural networks | |
| CN106845430A (en) | Pedestrian detection and tracking based on acceleration region convolutional neural networks | |
| CN112836639A (en) | Pedestrian multi-target tracking video identification method based on improved YOLOv3 model | |
| CN110633632A (en) | Weak supervision combined target detection and semantic segmentation method based on loop guidance | |
| CN109447082B (en) | Scene moving object segmentation method, system, storage medium and equipment | |
| CN107247956A (en) | A kind of fast target detection method judged based on grid | |
| CN107657625A (en) | Merge the unsupervised methods of video segmentation that space-time multiple features represent | |
| CN108648211A (en) | A kind of small target detecting method, device, equipment and medium based on deep learning | |
| CN110298297A (en) | Flame identification method and device | |
| Wan et al. | AFSar: An anchor-free SAR target detection algorithm based on multiscale enhancement representation learning | |
| CN110889421A (en) | Target detection method and device | |
| CN110390289A (en) | Based on the video security protection detection method for censuring understanding | |
| CN110956158A (en) | Pedestrian shielding re-identification method based on teacher and student learning frame | |
| CN114821152B (en) | Domain self-adaptive target detection method and system based on foreground-class perception alignment | |
| JP2022082493A (en) | Pedestrian re-identification method for random shielding recovery based on noise channel | |
| Wu et al. | Traffic sign detection based on SSD combined with receptive field module and path aggregation network | |
| Mijić et al. | Traffic sign detection using YOLOv3 | |
| CN110751005B (en) | Pedestrian detection method integrating depth perception features and kernel extreme learning machine | |
| Li et al. | An outstanding adaptive multi-feature fusion YOLOv3 algorithm for the small target detection in remote sensing images | |
| CN113129336A (en) | End-to-end multi-vehicle tracking method, system and computer readable medium | |
| CN116824333A (en) | Nasopharyngeal carcinoma detecting system based on deep learning model | |
| CN114067359B (en) | Pedestrian detection method integrating human body key points and visible part attention characteristics | |
| Wang et al. | Research on improved pedestrian detection algorithm based on convolutional neural network |
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: 20190430 |