CN107145845A - The pedestrian detection method merged based on deep learning and multi-characteristic points - Google Patents
The pedestrian detection method merged based on deep learning and multi-characteristic points Download PDFInfo
- Publication number
- CN107145845A CN107145845A CN201710280571.9A CN201710280571A CN107145845A CN 107145845 A CN107145845 A CN 107145845A CN 201710280571 A CN201710280571 A CN 201710280571A CN 107145845 A CN107145845 A CN 107145845A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- training
- mrow
- network
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/24765—Rule-based classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Abstract
The present invention relates to a kind of pedestrian detection method merged based on deep learning and multi-characteristic points, pedestrian image under training stage first acquisition applications scene simultaneously marks the head and shoulder position of pedestrian in image and then these pedestrian samples is used for into model training, and model training is divided into two steps:1) two disaggregated models at one pedestrian's head and shoulder position of training sample and training are used as using the head shoulder images of pedestrian;2) with step 1) train obtained model parameter to initialize the partial parameters of pedestrian detection model in the way of transfer learning.The present invention can overcome the problem of pedestrian is mutually blocked to a certain extent;Pedestrian's feature is extracted using deep learning method, the actual application problem that can preferably overcome the factors such as pedestrian's clothing, posture, residing background, illumination condition to change;Can also be efficiently against pedestrian's multi-pose, the problems such as pedestrian is multiple dimensioned and pedestrian is mutually blocked substantially increases the accuracy rate and robustness of pedestrian detection.
Description
Technical field
It is specifically that one kind is based on deep learning and multiple features the present invention relates to image procossing and technical field of computer vision
The pedestrian detection method of point fusion.
Background technology
Computer vision relate to image procossing, machine learning, pattern knowledge as an important branch of artificial intelligence
The cutting edge technology in many fields such as, do not automatically control.Computer vision is developed so far, be widely used in safety monitoring,
The various fields such as intelligent transportation, automatic Pilot, intelligent robot, industrial detection and space flight and aviation.
Target detection is the heat subject of computer vision field, wherein pedestrian detection be all for a long time academia and
The focus of attention of industrial quarters.Pedestrian detection technology is exactly for given image or video, it is necessary to allow computer automatic decision
Go out and whether there is pedestrian in image or video, also need to allow computer to determine all pedestrians in the picture specific if there is pedestrian
Position.Pedestrian detection is not an isolated technology, and it has with many problems closely contacts.Pedestrian detection be pedestrian with
Track, pedestrian's gait analysis, pedestrian behavior analysis, pedestrian's identity such as recognize at the important foundation and premise of application again.Therefore, Hang Renjian
Measuring tool has high scientific research and commercial value.
Conventional pedestrian detection method substantially has at present:Frame difference method, optical flow method, Background difference, the side based on template matches
Method and the method for feature based classification.Wherein, first four kinds be all based on image processing techniques pedestrian detection method.Such method
Although realizing simply, the factors such as pedestrian's posture, clothing, residing background, illumination condition are changed and pedestrian mutually hides
The processing of situations such as gear is less desirable.Therefore, the problem of there is low accuracy of detection and poor robustness in actual applications.
The pedestrian detection algorithm flow of feature based classification is as follows:It is the training stage first, extracts the feature of pedestrian, use these pedestrians
Feature takes the mode of machine learning to train two graders.Then in test phase, algorithm extracts test sample first
Feature, the feature extracted is input in grader to determine whether pedestrian.The method of feature based classification is by carrying
The high-level characteristic of image is taken, pedestrian's clothing, posture, residing background, illumination bar can be preferably overcome with reference to reliable grader
The application problem that the factors such as part change.Such pedestrian detection method turns into main flow algorithm.
The pedestrian detection method of feature based classification generally comprises pedestrian's feature extraction, classifier training, three ranks of detection
Section.In pedestrian's detection field, most common method is gradient orientation histogram (Histogram of Oriented
Gradient, HOG) combination supporting vector machine (Support Vector Machine, SVM) method.HOG juche idea is
The shape of localized target in image can well be represented by the density information at gradient or edge.Therefore, HOG features is main
Extraction process is exactly to calculate the gradient orientation histogram with the regional area of statistical picture.HOG feature combination SVM classifier methods
And other obtain relatively good testing result based on the pedestrian detection method that HOG features are improved.But pedestrian's multi-pose with
And larger defect is still had in the detection robustness mutually blocked of pedestrian.
The pedestrian detection method of current main flow have HOG features combination SVM classifier method and other changed based on HOG features
Good pedestrian detection method, such method describes pedestrian by extracting the gradient information of image, and combines machine learning neck
The grader in domain discriminates whether as pedestrian.In addition, as depth learning technology is obtained in image classification, field of image recognition
Immense success, depth learning technology has been also applied to pedestrian detection field.Wherein using RCNN detection frameworks as representative.Such side
Method first extracts candidate region and then candidate region is input into convolutional neural networks and discriminated whether to complete pedestrian for pedestrian
Detection task.
HOG characterization methods and other pedestrian spies that hand-designed is passed through based on the pedestrian detection method that HOG features are improved
The grader levied and combine machine learning areas is differentiated.Although such method can preferably overcome pedestrian's clothing, posture,
The actual application problem that the factors such as residing background, illumination condition change, but in pedestrian's multi-pose, pedestrian is multiple dimensioned and goes
Still without good treatment effect on the problem of people is mutually blocked.In addition, the current pedestrian detection side based on deep learning
Although method can overcome the problem of HOG characterization methods are present by learning the high-level characteristic of image, such method needs a large amount of
Calculate, be unfavorable for detection in real time..
The content of the invention
For the shortcoming of existing pedestrian detection method, the present invention proposes a kind of to merge based on deep learning and multi-characteristic points
Pedestrian detection method, can and pedestrian multiple dimensioned efficiently against pedestrian's multi-pose, pedestrian the problems such as mutually block, and adopt
With full convolutional network and the strategy of shared parameter, detection speed can be greatly improved.
Technical scheme is as follows to achieve these goals:
A kind of pedestrian detection method merged based on deep learning and multi-characteristic points, includes training stage and detection-phase.
Pedestrian image under training stage acquisition applications scene first and mark pedestrian in image head and shoulder position then will
These pedestrian samples are used for model training.Wherein model training is divided into two steps:1) head shoulder images using pedestrian are used as training
Sample simultaneously trains two disaggregated models at pedestrian's head and shoulder position by the way of Triplet Loss;2) step 1 is used) training
Obtained model parameter initializes the partial parameters of pedestrian detection model in the way of " transfer learning ".The pedestrian detection mould
The model that type is used as last detection-phase, employs training method end to end, contains candidate region extraction, Hang Rente
Levy the function of extraction and tagsort.
The pedestrian detection method merged based on deep learning and multi-characteristic points that the present invention is provided selects the head and shoulder portion of pedestrian
Position can overcome the problem of pedestrian is mutually blocked to a certain extent as detection target.In addition, the present invention uses deep learning
Method extracts pedestrian's feature.The factors such as pedestrian's clothing, posture, residing background, illumination condition not only can be preferably overcome to send out
The raw actual application problem changed, can also be efficiently against pedestrian's multi-pose, and pedestrian is multiple dimensioned and pedestrian is mutually blocked
Problem, substantially increases the accuracy rate and robustness of pedestrian detection.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, not
Inappropriate limitation of the present invention is constituted, in the accompanying drawings:
Fig. 1 is a kind of pedestrian's head and shoulder detection method entirety merged based on deep learning and multi-characteristic points disclosed by the invention
Flow chart;
The network structure that Fig. 2 uses for this bright pre-training step;
Fig. 3 is the Downsapling method (landmark-sensitive of distinguished point based zoning proposed by the present invention
Pooling the pedestrian's head and shoulder region division schematic diagram used in);
The network structure of Fig. 4 pedestrian detection models proposed by the present invention.
Embodiment
Describe the present invention in detail below in conjunction with accompanying drawing and specific embodiment, herein illustrative examples of the invention
And explanation is used for explaining the present invention, but it is not as a limitation of the invention.
Embodiment
A kind of pedestrian detection method merged based on deep learning and multi-characteristic points, includes training stage and detection-phase.
Pedestrian image under training stage acquisition applications scene first and mark pedestrian in image head and shoulder position then will
These pedestrian samples are used for model training.Wherein model training is divided into two steps:1) head shoulder images using pedestrian are used as training
Sample simultaneously trains two disaggregated models at pedestrian's head and shoulder position by the way of Triplet Loss;2) step 1 is used) training
Obtained model parameter initializes the partial parameters of pedestrian detection model in the way of " transfer learning ".The pedestrian detection mould
The model that type is used as last detection-phase, employs training method end to end, contains candidate region extraction, Hang Rente
Levy the function of extraction and tagsort.
Because the final detection model that the training stage obtains is a model end to end, detection-phase only need by
Image to be detected input pedestrian's detection model is that can obtain testing result.
The inventive method flow chart is as shown in Figure 1.
The emphasis of the present invention program is the training stage, and the training stage can be divided into three steps:1) gather and mark training number
According to collection;2) two disaggregated models are trained;3) the pedestrian detection network of training multi-characteristic points fusion.
Gather and mark training dataset
Gather practical application scene image.Then pedestrian's head and shoulder position in image is labeled using rectangle frame, and
The top left co-ordinate of rectangle frame in the picture and lower right corner coordinate record are got off, { x is designated as0, y0, x1, y1}.Finally will mark
Good image carries out flip horizontal and carrys out EDS extended data set, correspondingly, and the rectangle frame at mark pedestrian's head and shoulder position carries out flip horizontal
It is designated as { x '0, y '0, x '1, y '1}。
Train two disaggregated models
Using Triplet Loss object functions, two sorter networks are trained.Wherein convolution layer parameter is in follow-up migration
Used in habit as shared parameter.The purpose of the step is that training obtains low-level image feature, network is had preliminary differentiation pedestrian target
Ability.
Three sub-steps of pre-training point are carried out:
1st, the image-region at the pedestrian's head and shoulder position marked is cut out and is used as positive sample, while also cutting in right amount
Background image region into unifies size as negative sample, and by positive sample and the size adjusting of negative sample;
2nd, network structure is defined, the structure of network first half is consistent with pedestrian detection network, it is convenient to obtain pre-training
Network parameter move to pedestrian detection network, be then used as loss function using formula 1.The effect of formula 1 is to allow network
The positive sample feature of extraction and the distance between positive sample feature are smaller than positive sample feature and the distance between negative sample feature;
3rd, positive sample and negative sample are sequentially input to pre-training network according to (positive sample 1, positive sample 2, negative sample 1)
It is trained.
Pre-training network structure is as shown in Figure 2.
Train the pedestrian detection network of multi-characteristic points fusion
The model that the step is obtained using pre-training is initialized, and regard the whole pedestrian's scene image marked as instruction
White silk collects, and training obtains integrating complete pedestrian's detection model that candidate region is extracted and pedestrian classifies.
Pedestrian detection three sub-steps of model point are trained to carry out:
1st, using whole pedestrian's scene image as training image, with corresponding pedestrian's head and shoulder rectangle frame coordinate conduct in the lump
The input of network;
2nd, network structure is defined, whole network extracts sub-network (Region Proposal comprising candidate window
Network, RPN) and based on region full convolutional network (Region-Based Fully Convolutional Network,
RFCN).RPN effect is that the region for being suspected to be pedestrian is extracted from testing image, and specific practice is by last convolutional layer
Each position original image region is mapped back with the combination of three kinds of yardsticks and three kinds of length-width ratios, three kinds of yardsticks are respectively
[1282, 1282, 1282], three kinds of length-width ratios are respectively [1:1,1:2,2:1], the image-region mapped in this way substantially may be used
To cover all target areas of original image.RFCN effect is to differentiate whether the candidate region that RPN is provided is pedestrian and to row
The position of people is for further adjustments.The characteristic informations such as head, shoulder in order to make full use of pedestrian's head and shoulder position, the present invention is proposed
The Downsapling method of distinguished point based zoning (landmark-sensitive pooling) is to pedestrian's head and shoulder position
Multiple characteristic point informations melt being incorporated as distinguishing rule, and the dividing mode in pedestrian's head and shoulder region is as shown in Figure 3.While in order to
Improve arithmetic speed and keep higher Detection accuracy, the present invention uses convolutional layer to substitute full articulamentum as classification layer and seat
Mark returns layer.RPN and RFCN shares partial parameters layer, and complete network structure is as shown in Figure 4.
3rd, complete pedestrian detection model is obtained using the training of multitask training method end to end.
The technical scheme provided above the embodiment of the present invention is described in detail, specific case used herein
Principle and embodiment to the embodiment of the present invention are set forth, and the explanation of above example is only applicable to help and understands this
The principle of inventive embodiments;Simultaneously for those of ordinary skill in the art, according to the embodiment of the present invention, in specific embodiment party
It will change in formula and application, in summary, this specification content should not be construed as limiting the invention.
Claims (5)
1. a kind of pedestrian detection method merged based on deep learning and multi-characteristic points, comprising training stage and detection-phase, its
It is characterised by:
Pedestrian image under training stage acquisition applications scene first and mark pedestrian in image head and shoulder position then by these
Pedestrian sample is used for model training, and wherein model training is divided into two steps:
1) pedestrian's head and shoulder is trained using the head shoulder images of pedestrian as training sample and by the way of Triplet Loss
Two disaggregated models at position;
2) with step 1) train obtained model parameter to initialize the part of pedestrian detection model in the way of transfer learning
Parameter;
The model that the pedestrian detection model is used as last detection-phase, employs training method end to end, contains time
The function of favored area extraction, pedestrian's feature extraction and tagsort.
2. the pedestrian detection method as claimed in claim 1 merged based on deep learning and multi-characteristic points, it is characterised in that institute
Stating the training stage comprises the following steps:
1) gather and mark training dataset;
2) two disaggregated models are trained;
3) the pedestrian detection network of training multi-characteristic points fusion.
3. the pedestrian detection method as claimed in claim 2 merged based on deep learning and multi-characteristic points, it is characterised in that institute
State to gather and mark training dataset and comprise the following steps:
Practical application scene image is gathered, then pedestrian's head and shoulder position in image is labeled using rectangle frame, and by square
The top left co-ordinate of shape frame in the picture gets off with lower right corner coordinate record, is designated as { x0, y0, x1, y1, it will finally mark
Image carries out flip horizontal and carrys out EDS extended data set, correspondingly, and the rectangle frame at mark pedestrian's head and shoulder position carries out flip horizontal and is designated as
{x′0, y '0, x '1, y '1}。
4. the pedestrian detection method as claimed in claim 2 merged based on deep learning and multi-characteristic points, it is characterised in that institute
Two disaggregated models of training are stated to comprise the following steps:
1) image-region at the pedestrian's head and shoulder position marked is cut out and is used as positive sample, while also cutting appropriate background
Image-region into unifies size as negative sample, and by positive sample and the size adjusting of negative sample;
2) network structure is defined, the structure of network first half is consistent with pedestrian detection network, the convenient net for obtaining pre-training
Network parameter moves to pedestrian detection network and gone, and is then used as loss function using below equation:
<mrow>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mi>p</mi>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mi>p</mi>
<mn>2</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo>|</mo>
<mo>+</mo>
<mi>m</mi>
<mi>arg</mi>
<mi>i</mi>
<mi>n</mi>
<mo><</mo>
<mo>|</mo>
<mo>|</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mi>p</mi>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo>|</mo>
<mo>;</mo>
</mrow>
3) positive sample and negative sample are input into pre-training network in sequence to be trained.
5. the pedestrian detection method as claimed in claim 2 merged based on deep learning and multi-characteristic points, it is characterised in that institute
The pedestrian detection network for stating training multi-characteristic points fusion comprises the following steps:
1) whole pedestrian's scene image is used as training image, with corresponding pedestrian's head and shoulder rectangle frame coordinate in the lump as network
Input;
2) define network structure, whole network comprising candidate window extract sub-network (Region Proposal Network,
RPN the full convolutional network (Region-Based Fully Convolutional Network, RFCN)) and based on region;
3) complete pedestrian detection model is obtained using the training of multitask training method end to end.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710280571.9A CN107145845A (en) | 2017-04-26 | 2017-04-26 | The pedestrian detection method merged based on deep learning and multi-characteristic points |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710280571.9A CN107145845A (en) | 2017-04-26 | 2017-04-26 | The pedestrian detection method merged based on deep learning and multi-characteristic points |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107145845A true CN107145845A (en) | 2017-09-08 |
Family
ID=59773879
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710280571.9A Pending CN107145845A (en) | 2017-04-26 | 2017-04-26 | The pedestrian detection method merged based on deep learning and multi-characteristic points |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107145845A (en) |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107657249A (en) * | 2017-10-26 | 2018-02-02 | 珠海习悦信息技术有限公司 | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again |
CN107818302A (en) * | 2017-10-20 | 2018-03-20 | 中国科学院光电技术研究所 | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks |
CN108090472A (en) * | 2018-01-12 | 2018-05-29 | 浙江大学 | Pedestrian based on multichannel uniformity feature recognition methods and its system again |
CN108573279A (en) * | 2018-03-19 | 2018-09-25 | 精锐视觉智能科技(深圳)有限公司 | Image labeling method and terminal device |
CN108596211A (en) * | 2018-03-29 | 2018-09-28 | 中山大学 | It is a kind of that pedestrian's recognition methods again is blocked based on focusing study and depth e-learning |
CN108682003A (en) * | 2018-04-04 | 2018-10-19 | 睿视智觉(厦门)科技有限公司 | A kind of product quality detection method |
CN108805016A (en) * | 2018-04-27 | 2018-11-13 | 新智数字科技有限公司 | A kind of head and shoulder method for detecting area and device |
CN108921185A (en) * | 2018-05-04 | 2018-11-30 | 广州图匠数据科技有限公司 | A kind of shelf sales promotion information recognition methods based on image recognition, device and system |
CN109117888A (en) * | 2018-08-20 | 2019-01-01 | 北京旷视科技有限公司 | Recongnition of objects method and its neural network generation method and device |
CN109711316A (en) * | 2018-12-21 | 2019-05-03 | 广东工业大学 | A kind of pedestrian recognition methods, device, equipment and storage medium again |
CN110046632A (en) * | 2018-11-09 | 2019-07-23 | 阿里巴巴集团控股有限公司 | Model training method and device |
CN110069648A (en) * | 2017-09-25 | 2019-07-30 | 杭州海康威视数字技术股份有限公司 | A kind of image search method and device |
CN110070002A (en) * | 2019-03-29 | 2019-07-30 | 上海理工大学 | A kind of Activity recognition method based on 3D convolutional neural networks |
CN110096933A (en) * | 2018-01-30 | 2019-08-06 | 华为技术有限公司 | The method, apparatus and system of target detection |
CN110321818A (en) * | 2019-06-21 | 2019-10-11 | 江西洪都航空工业集团有限责任公司 | A kind of pedestrian detection method in complex scene |
WO2019196633A1 (en) * | 2018-04-10 | 2019-10-17 | 腾讯科技(深圳)有限公司 | Training method for image semantic segmentation model and server |
CN110378931A (en) * | 2019-07-10 | 2019-10-25 | 成都数之联科技有限公司 | A kind of pedestrian target motion track acquisition methods and system based on multi-cam |
CN110427920A (en) * | 2019-08-20 | 2019-11-08 | 武汉大学 | A kind of real-time pedestrian's analytic method towards monitoring environment |
CN112036446A (en) * | 2020-08-06 | 2020-12-04 | 汇纳科技股份有限公司 | Method, system, medium, and apparatus for target recognition feature fusion |
CN112132119A (en) * | 2020-11-24 | 2020-12-25 | 科大讯飞(苏州)科技有限公司 | Passenger flow statistical method and device, electronic equipment and storage medium |
CN112507904A (en) * | 2020-12-15 | 2021-03-16 | 重庆邮电大学 | Real-time classroom human body posture detection method based on multi-scale features |
CN112597943A (en) * | 2020-12-28 | 2021-04-02 | 北京眼神智能科技有限公司 | Feature extraction method and device for pedestrian re-identification, electronic equipment and storage medium |
CN112883880A (en) * | 2021-02-25 | 2021-06-01 | 电子科技大学 | Pedestrian attribute identification method based on human body structure multi-scale segmentation, storage medium and terminal |
CN116401961A (en) * | 2023-06-06 | 2023-07-07 | 广东电网有限责任公司梅州供电局 | Method, device, equipment and storage medium for determining pollution grade of insulator |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2804128A2 (en) * | 2013-03-22 | 2014-11-19 | MegaChips Corporation | Human detection device |
CN106203513A (en) * | 2016-07-08 | 2016-12-07 | 浙江工业大学 | A kind of based on pedestrian's head and shoulder multi-target detection and the statistical method of tracking |
CN106203506A (en) * | 2016-07-11 | 2016-12-07 | 上海凌科智能科技有限公司 | A kind of pedestrian detection method based on degree of depth learning art |
-
2017
- 2017-04-26 CN CN201710280571.9A patent/CN107145845A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2804128A2 (en) * | 2013-03-22 | 2014-11-19 | MegaChips Corporation | Human detection device |
CN106203513A (en) * | 2016-07-08 | 2016-12-07 | 浙江工业大学 | A kind of based on pedestrian's head and shoulder multi-target detection and the statistical method of tracking |
CN106203506A (en) * | 2016-07-11 | 2016-12-07 | 上海凌科智能科技有限公司 | A kind of pedestrian detection method based on degree of depth learning art |
Non-Patent Citations (2)
Title |
---|
DE CHENG: ""Person Re-identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function"", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
TIANRUI LIU: ""Fast head-shoulder proposal for deformable part model based pedestrian detection"", 《2016 IEEE INTERNATIONAL CONFERENCE ON DIGITAL SIGNAL PROCESSING》 * |
Cited By (38)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110069648A (en) * | 2017-09-25 | 2019-07-30 | 杭州海康威视数字技术股份有限公司 | A kind of image search method and device |
CN107818302A (en) * | 2017-10-20 | 2018-03-20 | 中国科学院光电技术研究所 | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks |
CN107657249A (en) * | 2017-10-26 | 2018-02-02 | 珠海习悦信息技术有限公司 | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again |
CN108090472A (en) * | 2018-01-12 | 2018-05-29 | 浙江大学 | Pedestrian based on multichannel uniformity feature recognition methods and its system again |
CN108090472B (en) * | 2018-01-12 | 2021-05-04 | 浙江大学 | Pedestrian re-identification method and system based on multi-channel consistency characteristics |
US11367272B2 (en) | 2018-01-30 | 2022-06-21 | Huawei Technologies Co., Ltd. | Target detection method, apparatus, and system |
WO2019149071A1 (en) * | 2018-01-30 | 2019-08-08 | 华为技术有限公司 | Target detection method, device, and system |
CN110096933A (en) * | 2018-01-30 | 2019-08-06 | 华为技术有限公司 | The method, apparatus and system of target detection |
CN108573279A (en) * | 2018-03-19 | 2018-09-25 | 精锐视觉智能科技(深圳)有限公司 | Image labeling method and terminal device |
CN108596211B (en) * | 2018-03-29 | 2020-08-28 | 中山大学 | Shielded pedestrian re-identification method based on centralized learning and deep network learning |
CN108596211A (en) * | 2018-03-29 | 2018-09-28 | 中山大学 | It is a kind of that pedestrian's recognition methods again is blocked based on focusing study and depth e-learning |
CN108682003A (en) * | 2018-04-04 | 2018-10-19 | 睿视智觉(厦门)科技有限公司 | A kind of product quality detection method |
US11348249B2 (en) | 2018-04-10 | 2022-05-31 | Tencent Technology (Shenzhen) Company Limited | Training method for image semantic segmentation model and server |
WO2019196633A1 (en) * | 2018-04-10 | 2019-10-17 | 腾讯科技(深圳)有限公司 | Training method for image semantic segmentation model and server |
CN108805016A (en) * | 2018-04-27 | 2018-11-13 | 新智数字科技有限公司 | A kind of head and shoulder method for detecting area and device |
CN108805016B (en) * | 2018-04-27 | 2022-02-08 | 新智数字科技有限公司 | Head and shoulder area detection method and device |
CN108921185A (en) * | 2018-05-04 | 2018-11-30 | 广州图匠数据科技有限公司 | A kind of shelf sales promotion information recognition methods based on image recognition, device and system |
CN109117888A (en) * | 2018-08-20 | 2019-01-01 | 北京旷视科技有限公司 | Recongnition of objects method and its neural network generation method and device |
CN110046632A (en) * | 2018-11-09 | 2019-07-23 | 阿里巴巴集团控股有限公司 | Model training method and device |
CN110046632B (en) * | 2018-11-09 | 2023-06-02 | 创新先进技术有限公司 | Model training method and device |
CN109711316A (en) * | 2018-12-21 | 2019-05-03 | 广东工业大学 | A kind of pedestrian recognition methods, device, equipment and storage medium again |
CN109711316B (en) * | 2018-12-21 | 2022-10-21 | 广东工业大学 | Pedestrian re-identification method, device, equipment and storage medium |
CN110070002A (en) * | 2019-03-29 | 2019-07-30 | 上海理工大学 | A kind of Activity recognition method based on 3D convolutional neural networks |
CN110321818A (en) * | 2019-06-21 | 2019-10-11 | 江西洪都航空工业集团有限责任公司 | A kind of pedestrian detection method in complex scene |
CN110378931A (en) * | 2019-07-10 | 2019-10-25 | 成都数之联科技有限公司 | A kind of pedestrian target motion track acquisition methods and system based on multi-cam |
CN110427920A (en) * | 2019-08-20 | 2019-11-08 | 武汉大学 | A kind of real-time pedestrian's analytic method towards monitoring environment |
CN110427920B (en) * | 2019-08-20 | 2021-11-02 | 武汉大学 | Real-time pedestrian analysis method oriented to monitoring environment |
CN112036446A (en) * | 2020-08-06 | 2020-12-04 | 汇纳科技股份有限公司 | Method, system, medium, and apparatus for target recognition feature fusion |
CN112036446B (en) * | 2020-08-06 | 2023-12-12 | 汇纳科技股份有限公司 | Method, system, medium and device for fusing target identification features |
CN112132119B (en) * | 2020-11-24 | 2021-05-28 | 科大讯飞(苏州)科技有限公司 | Passenger flow statistical method and device, electronic equipment and storage medium |
CN112132119A (en) * | 2020-11-24 | 2020-12-25 | 科大讯飞(苏州)科技有限公司 | Passenger flow statistical method and device, electronic equipment and storage medium |
CN112507904B (en) * | 2020-12-15 | 2022-06-03 | 重庆邮电大学 | Real-time classroom human body posture detection method based on multi-scale features |
CN112507904A (en) * | 2020-12-15 | 2021-03-16 | 重庆邮电大学 | Real-time classroom human body posture detection method based on multi-scale features |
CN112597943A (en) * | 2020-12-28 | 2021-04-02 | 北京眼神智能科技有限公司 | Feature extraction method and device for pedestrian re-identification, electronic equipment and storage medium |
CN112883880A (en) * | 2021-02-25 | 2021-06-01 | 电子科技大学 | Pedestrian attribute identification method based on human body structure multi-scale segmentation, storage medium and terminal |
CN112883880B (en) * | 2021-02-25 | 2022-08-19 | 电子科技大学 | Pedestrian attribute identification method based on human body structure multi-scale segmentation, storage medium and terminal |
CN116401961A (en) * | 2023-06-06 | 2023-07-07 | 广东电网有限责任公司梅州供电局 | Method, device, equipment and storage medium for determining pollution grade of insulator |
CN116401961B (en) * | 2023-06-06 | 2023-09-08 | 广东电网有限责任公司梅州供电局 | Method, device, equipment and storage medium for determining pollution grade of insulator |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107145845A (en) | The pedestrian detection method merged based on deep learning and multi-characteristic points | |
CN107168527B (en) | The first visual angle gesture identification and exchange method based on region convolutional neural networks | |
CN104463117B (en) | A kind of recognition of face sample collection method and system based on video mode | |
CN106709568B (en) | The object detection and semantic segmentation method of RGB-D image based on deep layer convolutional network | |
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
CN103810503B (en) | Depth study based method for detecting salient regions in natural image | |
CN107832672A (en) | A kind of pedestrian's recognition methods again that more loss functions are designed using attitude information | |
CN102708361B (en) | Human face collecting method at a distance | |
CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
CN108256421A (en) | A kind of dynamic gesture sequence real-time identification method, system and device | |
CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
CN107808143A (en) | Dynamic gesture identification method based on computer vision | |
CN103902962B (en) | One kind is blocked or the adaptive face identification method of light source and device | |
CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
CN105354565A (en) | Full convolution network based facial feature positioning and distinguishing method and system | |
CN107330453A (en) | The Pornographic image recognizing method of key position detection is recognized and merged based on substep | |
CN107133569A (en) | The many granularity mask methods of monitor video based on extensive Multi-label learning | |
CN107066916A (en) | Scene Semantics dividing method based on deconvolution neutral net | |
WO2009123354A1 (en) | Method, apparatus, and program for detecting object | |
JP7142420B2 (en) | Image processing device, learning method, trained model, image processing method | |
CN109558810A (en) | Divided based on position and merges target person recognition methods | |
CN107909081A (en) | The quick obtaining and quick calibrating method of image data set in a kind of deep learning | |
CN110298297A (en) | Flame identification method and device | |
CN106529494A (en) | Human face recognition method based on multi-camera model | |
CN110263768A (en) | A kind of face identification method based on depth residual error 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 |
Application publication date: 20170908 |
|
RJ01 | Rejection of invention patent application after publication |