CN108229445A - A kind of more people's Attitude estimation methods based on cascade pyramid network - Google Patents

A kind of more people's Attitude estimation methods based on cascade pyramid network Download PDF

Info

Publication number
CN108229445A
CN108229445A CN201810132802.6A CN201810132802A CN108229445A CN 108229445 A CN108229445 A CN 108229445A CN 201810132802 A CN201810132802 A CN 201810132802A CN 108229445 A CN108229445 A CN 108229445A
Authority
CN
China
Prior art keywords
network
key point
people
bounding box
cascade
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.)
Withdrawn
Application number
CN201810132802.6A
Other languages
Chinese (zh)
Inventor
夏春秋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Vision Technology Co Ltd
Original Assignee
Shenzhen Vision Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Vision Technology Co Ltd filed Critical Shenzhen Vision Technology Co Ltd
Priority to CN201810132802.6A priority Critical patent/CN108229445A/en
Publication of CN108229445A publication Critical patent/CN108229445A/en
Withdrawn legal-status Critical Current

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Image Analysis (AREA)

Abstract

A kind of more people's Attitude estimation methods based on cascade pyramid network proposed in the present invention, main contents include：Cascade pyramid network (CPN), more people's Attitude estimations, training and test, its process is, bounding box suggestion is first generated according to the anchor point of acquiescence, then it is cut according to characteristic pattern, and pass through recursive convolution neural network (R CNN) further refinement suggestion, to obtain final bounding box, bounding box irises out the personage in picture, then with cascade pyramid network key point is positioned in the bounding box of each personage, wherein global network can position simple key point, network is refined by integrating the character representation from all ranks of global network to handle difficult key point, and it is only lost from the key point backpropagation of selection.Method from up to down for more people's pose estimations, using pyramid network is cascaded, is substantially increased the performance of Attitude estimation, can adapt to the high request of Attitude estimation in practical application by the present invention.

Description

A kind of more people's Attitude estimation methods based on cascade pyramid network
Technical field
The present invention relates to Attitude estimation fields, estimate more particularly, to a kind of more people's postures based on cascade pyramid network Meter method.
Background technology
More people's Attitude estimations are that the key point of all persons in image is identified and positioned, it is human action's identification The challenge subjects that the basic research subject and tool of vision applications a variety of with human-computer interaction etc. acquire a certain degree of difficulty.More people's postures Estimation technique, which can be used for sports or nautch etc., needs the field estimated personage's posture, by sportsman Or the posture of performer is identified and analyzed, them can be helped to carry out objective and amount to the movement posture of oneself or other people The analysis of change or statistical correlation data, for creating personalized training and analysis system, instruct sportsman or performer into The scientific and effective training of row；It can be used for pedestrian's Attitude estimation of field of traffic, be carried out by the posture to numerous pedestrians Identification and analysis judge the direction that pedestrian advances, so as to assist driver's planning travelling line and take corresponding measure.It is relevant Attitude estimation technology can be also used for human-computer interaction, public arena the fields such as safety-protection system, brought more to people’s lives Facility.However, existing Attitude estimation method cannot still well solve due to block key point, stealthy key point with And the problem of accuracy is not high is estimated caused by complicated background.
The present invention proposes a kind of more people's Attitude estimation methods based on cascade pyramid network, first according to the anchor point of acquiescence Bounding box suggestion is generated, is then cut according to characteristic pattern, and passes through recursive convolution neural network (R-CNN) and further refines It is recommended that obtain final bounding box, bounding box irises out the personage in picture, then with cascade pyramid network in each personage Bounding box in position key point, wherein global network can position simple key point, refine network by integrating from complete The character representation of all ranks of office network handles difficult key point, and only lost from the key point backpropagation of selection.This hair It is bright that method from up to down is used for more people's pose estimations, using pyramid network is cascaded, substantially increase the property of Attitude estimation Energy can adapt to the high request of Attitude estimation in practical application.
Invention content
Estimate that accuracy is not high caused by for the key point due to blocking, stealthy key point and complicated background Problem, the purpose of the present invention is to provide a kind of more people's Attitude estimation methods based on cascade pyramid network, first according to acquiescence Anchor point generation bounding box suggestion, then cut according to characteristic pattern, and pass through recursive convolution neural network (R-CNN) into one Step refinement suggests that, to obtain final bounding box, bounding box irises out the personage in picture, then with cascade pyramid network every Key point is positioned in the bounding box of a personage, wherein global network can position simple key point, refine network and pass through integration Character representation from all ranks of global network handles difficult key point, and only damaged from the key point backpropagation of selection It loses.
To solve the above problems, the present invention provides a kind of more people's Attitude estimation methods based on cascade pyramid network, Main contents include：
(1) cascade pyramid network (CPN)；
(2) more people's Attitude estimations；
(3) training and test.
Wherein, the cascade pyramid network (CPN), cascade pyramid network include two sub-networks, respectively entirely Office network and refinement network；Global network is a feature pyramid network, can position the key point of " simple ", such as eyes and Hand, but possibly can not accurately identify and be blocked or sightless key point；Network is refined by integrating from global network to own The character representation of rank handles " hardly possible " key point.
Further, the global network, the last one by different convolution feature second to the 5th convolutional layer are residual Poor block is expressed as C2,C3,…,C5；In C2,C3,…,C5It is upper that 3 × 3 convolution filters is applied to generate the thermal map of key point； Such as shallow-layer feature C2And C3With higher spatial resolution, but the semantic information identified is less；And further feature layer C4And C5By In convolution (He Chihua) with more semantic informations, but spatial resolution is relatively low；Therefore, usually U-shaped structure is integrated, from And keep the spatial resolution and semantic information of characteristic layer；Feature pyramid network (FPN) is further by depth supervision message U-shaped structure is improved, similar feature pyramid structure is applied to crucial point estimation；Each element in upsampling process 1 × 1 convolution kernel is applied before summation process.
Further, the refinement network in order to improve the efficiency of information transmission and keep integrality, refines network and leads to The information for crossing different stage is transmitted, finally by up-sampling and cascade mode by the information integration of different levels；It refines All pyramid features are together in series by network；In addition, more bottleneck blocks are added in deeper level, smaller sky Between size good balance is achieved between efficiency；
With the continuous training of network, network often increasingly focuses on most of " simple " key point, and thinks little of hiding Gear and key point；Therefore in network is refined, the key point of " hardly possible " is clearly selected based on training loss, and only from the pass of selection The backpropagation of key point is lost.
Wherein, more people's Attitude estimations, the methods of more people's Attitude estimations are broadly divided into from bottom to top and from top to bottom Method；This method employs top-to-bottom method, i.e., positions first from image and iris out all persons with bounding box, so The single Attitude estimation in bounding box is solved the problems, such as afterwards；
If in order to obtain good performance, then personage is needed to examine for more people's pose estimations method from up to down Survey device and single pose estimation device.
Further, the person detecting, detection method is usually made of two stages, first according to the anchor point of acquiescence Bounding box suggestion is generated, is then cut according to characteristic pattern, and passes through recursive convolution neural network (R-CNN) and further refines It is recommended that obtain final bounding box.
Further, the cutting, for the detection block of each personage, which is extended to one fixed high wide Than, such as height:Width=256:192, then from image cropping without warp image the ratio of width to height；Finally, by the figure after cutting The size of picture is adjusted to the fixed size of 256 pixel of default height and 192 pixels.
Wherein, the training and test, are verified on the data set comprising 5000 images, and test set includes surveying Try development set (20K images) and test challenge collection (20K images)；Most of experiments are all in object key point similarity (OKS) On the basis of assessed, wherein OKS defines the similarity between different human body posture；After image cropping, using random Overturning, Random-Rotation (- 40 °~+40 °) and random size (0.7~1.3) enhance data.
Further, the training, all Attitude estimation models are trained using stochastic gradient descent algorithm, Initial learning rate is 5 × 10-4；Learning rate every 10 periods reduce by 2 times；Use 10-5Weight attenuation, and make in a network It is normalized with batch.
Further, the test during test, in order to minimize the variance of prediction, applies two on the hot spot of prediction Tie up Gaussian filter；It predicts the posture of corresponding flipped image and the thermal map that is averaged is finally to be predicted；It is responded using from highest A quarter to the second high response direction deviates the final position to obtain key point.
Description of the drawings
Fig. 1 is a kind of system framework figure of more people's Attitude estimation methods based on cascade pyramid network of the present invention.
Fig. 2 is a kind of cascade pyramid network knot of more people's Attitude estimation methods based on cascade pyramid network of the present invention Structure.
Fig. 3 is a kind of heat outputting of the different characteristic of more people's Attitude estimation methods based on cascade pyramid network of the present invention Figure.
Specific embodiment
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the application can phase It mutually combines, the present invention is described in further detail in the following with reference to the drawings and specific embodiments.
Fig. 1 is a kind of system framework figure of more people's Attitude estimation methods based on cascade pyramid network of the present invention.Mainly Including cascade pyramid network (CPN)；More people's Attitude estimations；Training and test.
The method of more people's Attitude estimations is broadly divided into from bottom to top and top-to-bottom method；This method employ from upper and Under method, i.e., first from image position and iris out all persons with bounding box, then solve bounding box in single posture Estimation problem；
If in order to obtain good performance, then personage is needed to examine for more people's pose estimations method from up to down Survey device and single pose estimation device.
Detection method is usually made of two stages, generates bounding box suggestion according to the anchor point of acquiescence first, then basis Characteristic pattern is cut, and passes through recursive convolution neural network (R-CNN) further refinement suggestion, to obtain final boundary Frame.
For the detection block of each personage, which is extended to a fixed depth-width ratio, such as height:Width=256: 192, then from image cropping without warp image the ratio of width to height；Finally, the size of the image after cutting is adjusted to default height The fixed size of 256 pixels and 192 pixels.
Training and test are verified, test set includes test development collection on the data set comprising 5000 images (20K images) and test challenge collection (20K images)；Most of experiments are all enterprising on object key point similarity (OKS) basis Row assessment, wherein OKS defines the similarity between different human body posture；After image cropping, using random overturning, at random (- 40 °~+40 °) and random size (0.7~1.3) are rotated to enhance data.
All Attitude estimation models are trained using stochastic gradient descent algorithm, and initial learning rate is 5 × 10-4；Learning rate every 10 periods reduce by 2 times；Use 10-5Weight attenuation, and in a network using batch normalize.
During test, in order to minimize the variance of prediction, 2-d gaussian filters device is applied on the hot spot of prediction；Prediction is corresponding Flipped image posture and the thermal map that is averaged finally to be predicted；Four points of the second high response direction are responsive to using from highest One of offset obtain the final position of key point.
Fig. 2 is a kind of cascade pyramid network knot of more people's Attitude estimation methods based on cascade pyramid network of the present invention Structure.It cascades pyramid network and includes two sub-networks, respectively global network and refinement network；Global network is a feature gold Word tower network can position the key point of " simple ", such as eyes and hand, but possibly can not accurately identify and be blocked or sightless Key point；Network is refined by integrating the character representation from all ranks of global network to handle " hardly possible " key point.
Fig. 3 is a kind of heat outputting of the different characteristic of more people's Attitude estimation methods based on cascade pyramid network of the present invention Figure.As shown in figure 3, global network can efficiently locate the key point of eyes, but it possibly can not be accurately positioned the position of buttocks It puts；The positioning of the key point as buttocks usually requires more contextual informations rather than neighbouring external appearance characteristic；Cause This, based on global network Direct Recognition, these " hard " key points are often difficult, it is therefore desirable to refine network to handle this Problem.
Wherein, the global network, by the last one residual block of different convolution feature second to the 5th convolutional layer It is expressed as C2,C3,…,C5；In C2,C3,…,C5It is upper that 3 × 3 convolution filters is applied to generate the thermal map of key point；It is such as shallow Layer feature C2And C3With higher spatial resolution, but the semantic information identified is less；And further feature layer C4And C5Due to volume It accumulates (He Chihua) and there are more semantic informations, but spatial resolution is relatively low；Therefore, usually U-shaped structure is integrated, so as to protect Hold the spatial resolution and semantic information of characteristic layer；Feature pyramid network (FPN) is further improved by depth supervision message Similar feature pyramid structure is applied to crucial point estimation by U-shaped structure；Each element summation in upsampling process 1 × 1 convolution kernel is applied before process.
Wherein, the refinement network in order to improve the efficiency of information transmission and keep integrality, refines network and passes through not The information of same level is transmitted, finally by up-sampling and cascade mode by the information integration of different levels；Refine network All pyramid features are together in series；In addition, more bottleneck blocks are added in deeper level, smaller space ruler It is very little that good balance is achieved between efficiency；
With the continuous training of network, network often increasingly focuses on most of " simple " key point, and thinks little of hiding Gear and key point；Therefore in network is refined, the key point of " hardly possible " is clearly selected based on training loss, and only from the pass of selection The backpropagation of key point is lost.
For those skilled in the art, the present invention is not limited to the details of above-described embodiment, in the essence without departing substantially from the present invention In the case of refreshing and range, the present invention can be realized in other specific forms.In addition, those skilled in the art can be to this hair Bright to carry out various modification and variations without departing from the spirit and scope of the present invention, these improvements and modifications also should be regarded as the present invention's Protection domain.Therefore, appended claims are intended to be construed to include preferred embodiment and fall into all changes of the scope of the invention More and change.

Claims (10)

  1. A kind of 1. more people's Attitude estimation methods based on cascade pyramid network, which is characterized in that main to include cascade pyramid Network (CPN) (one)；More people's Attitude estimations (two)；Training and test (three).
  2. 2. based on the cascade pyramid network (CPN) (one) described in claims 1, which is characterized in that cascade pyramid network Including two sub-networks, respectively global network and refinement network；Global network is a feature pyramid network, can be positioned The key point of " simple " such as eyes and hand, but possibly can not be accurately identified and be blocked or sightless key point；Network is refined to lead to It crosses and integrates the character representation from all ranks of global network to handle " hardly possible " key point.
  3. 3. based on the global network described in claims 2, which is characterized in that by different convolution feature second to the 5th convolution The last one residual block of layer is expressed as C2,C3,…,C5；In C2,C3,…,C5It is upper that 3 × 3 convolution filters is applied to generate The thermal map of key point；Such as shallow-layer feature C2And C3With higher spatial resolution, but the semantic information identified is less；And deep layer Characteristic layer C4And C5There are more semantic informations due to convolution (He Chihua), but spatial resolution is relatively low；Therefore, usually by U Shape structural integrity, so as to keep the spatial resolution of characteristic layer and semantic information；Feature pyramid network (FPN) is supervised by depth It superintends and directs information and further improves U-shaped structure, similar feature pyramid structure is applied to crucial point estimation；In upsampling process In each element summation process before apply 1 × 1 convolution kernel.
  4. 4. based on the refinement network described in claims 2, which is characterized in that in order to improve the efficiency of information transmission and keep Whole property is refined network and is transmitted by the information of different stage, finally by up-sampling and cascade mode by different levels Information integration；It refines network all pyramid features are together in series；In addition, more bottleneck blocks are added to deeper layer In secondary, smaller bulk achieves good balance between efficiency；
    With the continuous training of network, network often increasingly focuses on most of " simple " key point, and think little of blocking with Key point；Therefore in network is refined, the key point of " hardly possible " is clearly selected based on training loss, and only from the key point of selection Backpropagation is lost.
  5. 5. more people's Attitude estimations (two) described in based on claims 1, which is characterized in that the method for more people's Attitude estimations is main It is divided into from bottom to top and top-to-bottom method；This method employs top-to-bottom method, i.e., first from image positioning and All persons are irised out with bounding box, then solve the problems, such as the single Attitude estimation in bounding box；
    If method from up to down in order to obtain good performance, is then needed into person detector for more people's pose estimations And single pose estimation device.
  6. 6. based on the person detecting described in claims 5, which is characterized in that detection method is usually made of two stages, first Bounding box suggestion is first generated according to the anchor point of acquiescence, is then cut according to characteristic pattern, and pass through recursive convolution neural network (R-CNN) further refinement is suggested, to obtain final bounding box.
  7. 7. based on the cutting described in claims 6, which is characterized in that for the detection block of each personage, which is extended to One fixed depth-width ratio, such as height:Width=256:192, then from image cropping without warp image the ratio of width to height；Most Afterwards, the size of the image after cutting is adjusted to the fixed size of 256 pixel of default height and 192 pixels.
  8. 8. based on the training described in claims 1 and test (three), which is characterized in that in the data set for including 5000 images On verified, test set include test development collection (20K images) and test challenge collect (20K images)；Most of experiments are all It is assessed on the basis of object key point similarity (OKS), wherein OKS defines similar between different human body posture Degree；After image cropping, enhance number using random overturning, Random-Rotation (- 40 °~+40 °) and random size (0.7~1.3) According to.
  9. 9. based on the training described in claims 8, which is characterized in that all Attitude estimation models are all to use stochastic gradient Descent algorithm training, initial learning rate is 5 × 10-4；Learning rate every 10 periods reduce by 2 times；Use 10-5Weight decline Subtract, and normalized in a network using batch.
  10. 10. based on the test described in claims 8, which is characterized in that during test, in order to minimize the variance of prediction, pre- 2-d gaussian filters device is applied on the hot spot of survey；It predicts the posture of corresponding flipped image and the thermal map that is averaged is final pre- to obtain It surveys；The final position to obtain key point is deviated using a quarter that the second high response direction is responsive to from highest.
CN201810132802.6A 2018-02-09 2018-02-09 A kind of more people's Attitude estimation methods based on cascade pyramid network Withdrawn CN108229445A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810132802.6A CN108229445A (en) 2018-02-09 2018-02-09 A kind of more people's Attitude estimation methods based on cascade pyramid network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810132802.6A CN108229445A (en) 2018-02-09 2018-02-09 A kind of more people's Attitude estimation methods based on cascade pyramid network

Publications (1)

Publication Number Publication Date
CN108229445A true CN108229445A (en) 2018-06-29

Family

ID=62661331

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810132802.6A Withdrawn CN108229445A (en) 2018-02-09 2018-02-09 A kind of more people's Attitude estimation methods based on cascade pyramid network

Country Status (1)

Country Link
CN (1) CN108229445A (en)

Cited By (48)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108710830A (en) * 2018-04-20 2018-10-26 浙江工商大学 A kind of intensive human body 3D posture estimation methods for connecting attention pyramid residual error network and equidistantly limiting of combination
CN109345504A (en) * 2018-08-07 2019-02-15 浙江大学 A kind of bottom-up more people's Attitude estimation methods constrained using bounding box
CN109376571A (en) * 2018-08-03 2019-02-22 西安电子科技大学 Estimation method of human posture based on deformation convolution
CN109376681A (en) * 2018-11-06 2019-02-22 广东工业大学 A kind of more people's Attitude estimation method and system
CN109448007A (en) * 2018-11-02 2019-03-08 北京迈格威科技有限公司 Image processing method, image processing apparatus and storage medium
CN109472289A (en) * 2018-10-09 2019-03-15 北京陌上花科技有限公司 Critical point detection method and apparatus
CN109508681A (en) * 2018-11-20 2019-03-22 北京京东尚科信息技术有限公司 The method and apparatus for generating human body critical point detection model
CN109543549A (en) * 2018-10-26 2019-03-29 北京陌上花科技有限公司 Image processing method and device, mobile end equipment, server for more people's Attitude estimations
CN109658412A (en) * 2018-11-30 2019-04-19 湖南视比特机器人有限公司 It is a kind of towards de-stacking sorting packing case quickly identify dividing method
CN109711273A (en) * 2018-12-04 2019-05-03 北京字节跳动网络技术有限公司 Image key points extracting method, device, readable storage medium storing program for executing and electronic equipment
CN109784350A (en) * 2018-12-29 2019-05-21 天津大学 In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network
CN109815901A (en) * 2019-01-24 2019-05-28 杭州电子科技大学 A kind of more people's Attitude estimation methods based on YOLOv3 algorithm
CN109858430A (en) * 2019-01-28 2019-06-07 杭州电子科技大学 A kind of more people's attitude detecting methods based on intensified learning optimization
CN109948453A (en) * 2019-02-25 2019-06-28 华中科技大学 A kind of more people's Attitude estimation methods based on convolutional neural networks
CN110135375A (en) * 2019-05-20 2019-08-16 中国科学院宁波材料技术与工程研究所 More people's Attitude estimation methods based on global information integration
CN110163059A (en) * 2018-10-30 2019-08-23 腾讯科技（深圳）有限公司 More people's gesture recognition methods, device and electronic equipment
CN110163157A (en) * 2019-05-24 2019-08-23 南京邮电大学 A method of more people's Attitude estimations are carried out using novel loss function
CN110175575A (en) * 2019-05-29 2019-08-27 南京邮电大学 A kind of single Attitude estimation method based on novel high-resolution network model
CN110210402A (en) * 2019-06-03 2019-09-06 北京卡路里信息技术有限公司 Feature extracting method, device, terminal device and storage medium
CN110246181A (en) * 2019-05-24 2019-09-17 华中科技大学 Attitude estimation model training method, Attitude estimation method and system based on anchor point
CN110276316A (en) * 2019-06-26 2019-09-24 电子科技大学 A kind of human body critical point detection method based on deep learning
CN110321795A (en) * 2019-05-24 2019-10-11 平安科技（深圳）有限公司 User's gesture recognition method, device, computer installation and computer storage medium
CN110349164A (en) * 2019-07-19 2019-10-18 北京华捷艾米科技有限公司 A kind of image, semantic dividing method, device and terminal device
CN110378253A (en) * 2019-07-01 2019-10-25 浙江大学 A kind of real time critical point detecting method based on lightweight neural network
CN110443144A (en) * 2019-07-09 2019-11-12 天津中科智能识别产业技术研究院有限公司 A kind of human body image key point Attitude estimation method
CN110889858A (en) * 2019-12-03 2020-03-17 中国太平洋保险（集团）股份有限公司 Automobile part segmentation method and device based on point regression
CN110929638A (en) * 2019-11-20 2020-03-27 北京奇艺世纪科技有限公司 Human body key point identification method and device and electronic equipment
CN110942056A (en) * 2018-09-21 2020-03-31 深圳云天励飞技术有限公司 Clothing key point positioning method and device, electronic equipment and medium
WO2020062493A1 (en) * 2018-09-29 2020-04-02 北京字节跳动网络技术有限公司 Image processing method and apparatus
CN110969087A (en) * 2019-10-31 2020-04-07 浙江省北大信息技术高等研究院 Gait recognition method and system
CN111062261A (en) * 2019-11-25 2020-04-24 维沃移动通信（杭州）有限公司 Image processing method and device
CN111107278A (en) * 2018-10-26 2020-05-05 北京微播视界科技有限公司 Image processing method and device, electronic equipment and readable storage medium
CN111104841A (en) * 2019-09-16 2020-05-05 平安科技（深圳）有限公司 Violent behavior detection method and system
CN111291716A (en) * 2020-02-28 2020-06-16 深圳大学 Sperm cell recognition method, device, computer equipment and storage medium
CN111695519A (en) * 2020-06-12 2020-09-22 北京百度网讯科技有限公司 Key point positioning method, device, equipment and storage medium
CN112036244A (en) * 2020-07-30 2020-12-04 广东技术师范大学 Human body posture estimation method based on neural network
CN112131959A (en) * 2020-08-28 2020-12-25 浙江工业大学 2D human body posture estimation method based on multi-scale feature reinforcement
CN112418046A (en) * 2020-11-17 2021-02-26 武汉云极智能科技有限公司 Fitness guidance method, storage medium and system based on cloud robot
CN112597955A (en) * 2020-12-30 2021-04-02 华侨大学 Single-stage multi-person attitude estimation method based on feature pyramid network
CN112651294A (en) * 2020-11-05 2021-04-13 同济大学 Method for recognizing human body shielding posture based on multi-scale fusion
CN112966574A (en) * 2021-02-22 2021-06-15 厦门艾地运动科技有限公司 Human body three-dimensional key point prediction method and device and electronic equipment
CN113033524A (en) * 2021-05-26 2021-06-25 北京的卢深视科技有限公司 Occlusion prediction model training method and device, electronic equipment and storage medium
TWI733616B (en) * 2020-11-04 2021-07-11 財團法人資訊工業策進會 Reconition system of human body posture, reconition method of human body posture, and non-transitory computer readable storage medium
CN113221626A (en) * 2021-03-04 2021-08-06 北京联合大学 Human body posture estimation method based on Non-local high-resolution network
CN113569798A (en) * 2018-11-16 2021-10-29 北京市商汤科技开发有限公司 Key point detection method and device, electronic equipment and storage medium
US11270147B1 (en) 2020-10-05 2022-03-08 International Business Machines Corporation Action-object recognition in cluttered video scenes using text
US11423252B1 (en) 2021-04-29 2022-08-23 International Business Machines Corporation Object dataset creation or modification using labeled action-object videos
CN117392761A (en) * 2023-12-13 2024-01-12 深圳须弥云图空间科技有限公司 Human body pose recognition method and device, electronic equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106529374A (en) * 2015-09-10 2017-03-22 大唐电信科技股份有限公司 Cascaded face key point positioning method and system
CN107038429A (en) * 2017-05-03 2017-08-11 四川云图睿视科技有限公司 A kind of multitask cascade face alignment method based on deep learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106529374A (en) * 2015-09-10 2017-03-22 大唐电信科技股份有限公司 Cascaded face key point positioning method and system
CN107038429A (en) * 2017-05-03 2017-08-11 四川云图睿视科技有限公司 A kind of multitask cascade face alignment method based on deep learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YILUN CHEN，ZHICHENG WANG,ET.AL.: ""Cascaded Pyramid Network for Multi-Person Pose Estimation"", 《ARXIV》 *

Cited By (77)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108710830A (en) * 2018-04-20 2018-10-26 浙江工商大学 A kind of intensive human body 3D posture estimation methods for connecting attention pyramid residual error network and equidistantly limiting of combination
CN108710830B (en) * 2018-04-20 2020-08-28 浙江工商大学 Human body 3D posture estimation method combining dense connection attention pyramid residual error network and isometric limitation
CN109376571A (en) * 2018-08-03 2019-02-22 西安电子科技大学 Estimation method of human posture based on deformation convolution
CN109376571B (en) * 2018-08-03 2022-04-08 西安电子科技大学 Human body posture estimation method based on deformation convolution
CN109345504A (en) * 2018-08-07 2019-02-15 浙江大学 A kind of bottom-up more people's Attitude estimation methods constrained using bounding box
CN110942056A (en) * 2018-09-21 2020-03-31 深圳云天励飞技术有限公司 Clothing key point positioning method and device, electronic equipment and medium
WO2020062493A1 (en) * 2018-09-29 2020-04-02 北京字节跳动网络技术有限公司 Image processing method and apparatus
CN109472289B (en) * 2018-10-09 2022-03-29 北京陌上花科技有限公司 Key point detection method and device
CN109472289A (en) * 2018-10-09 2019-03-15 北京陌上花科技有限公司 Critical point detection method and apparatus
CN109543549A (en) * 2018-10-26 2019-03-29 北京陌上花科技有限公司 Image processing method and device, mobile end equipment, server for more people's Attitude estimations
CN111107278A (en) * 2018-10-26 2020-05-05 北京微播视界科技有限公司 Image processing method and device, electronic equipment and readable storage medium
CN109543549B (en) * 2018-10-26 2021-09-07 北京陌上花科技有限公司 Image data processing method and device for multi-person posture estimation, mobile terminal equipment and server
WO2020088433A1 (en) * 2018-10-30 2020-05-07 腾讯科技（深圳）有限公司 Method and apparatus for recognizing postures of multiple persons, electronic device, and storage medium
CN110163059A (en) * 2018-10-30 2019-08-23 腾讯科技（深圳）有限公司 More people's gesture recognition methods, device and electronic equipment
US11501574B2 (en) 2018-10-30 2022-11-15 Tencent Technology (Shenzhen) Company Limited Multi-person pose recognition method and apparatus, electronic device, and storage medium
CN110163059B (en) * 2018-10-30 2022-08-23 腾讯科技（深圳）有限公司 Multi-person posture recognition method and device and electronic equipment
CN109448007B (en) * 2018-11-02 2020-10-09 北京迈格威科技有限公司 Image processing method, image processing apparatus, and storage medium
CN109448007A (en) * 2018-11-02 2019-03-08 北京迈格威科技有限公司 Image processing method, image processing apparatus and storage medium
CN109376681B (en) * 2018-11-06 2021-09-03 广东工业大学 Multi-person posture estimation method and system
CN109376681A (en) * 2018-11-06 2019-02-22 广东工业大学 A kind of more people's Attitude estimation method and system
CN113569798A (en) * 2018-11-16 2021-10-29 北京市商汤科技开发有限公司 Key point detection method and device, electronic equipment and storage medium
CN109508681B (en) * 2018-11-20 2021-11-30 北京京东尚科信息技术有限公司 Method and device for generating human body key point detection model
CN109508681A (en) * 2018-11-20 2019-03-22 北京京东尚科信息技术有限公司 The method and apparatus for generating human body critical point detection model
CN109658412A (en) * 2018-11-30 2019-04-19 湖南视比特机器人有限公司 It is a kind of towards de-stacking sorting packing case quickly identify dividing method
CN109711273A (en) * 2018-12-04 2019-05-03 北京字节跳动网络技术有限公司 Image key points extracting method, device, readable storage medium storing program for executing and electronic equipment
CN109784350A (en) * 2018-12-29 2019-05-21 天津大学 In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network
CN109815901A (en) * 2019-01-24 2019-05-28 杭州电子科技大学 A kind of more people's Attitude estimation methods based on YOLOv3 algorithm
CN109858430A (en) * 2019-01-28 2019-06-07 杭州电子科技大学 A kind of more people's attitude detecting methods based on intensified learning optimization
CN109948453A (en) * 2019-02-25 2019-06-28 华中科技大学 A kind of more people's Attitude estimation methods based on convolutional neural networks
CN110135375A (en) * 2019-05-20 2019-08-16 中国科学院宁波材料技术与工程研究所 More people's Attitude estimation methods based on global information integration
CN110321795A (en) * 2019-05-24 2019-10-11 平安科技（深圳）有限公司 User's gesture recognition method, device, computer installation and computer storage medium
CN110321795B (en) * 2019-05-24 2024-02-23 平安科技（深圳）有限公司 User gesture recognition method and device, computer device and computer storage medium
CN110246181A (en) * 2019-05-24 2019-09-17 华中科技大学 Attitude estimation model training method, Attitude estimation method and system based on anchor point
CN110163157A (en) * 2019-05-24 2019-08-23 南京邮电大学 A method of more people's Attitude estimations are carried out using novel loss function
CN110246181B (en) * 2019-05-24 2021-02-26 华中科技大学 Anchor point-based attitude estimation model training method, attitude estimation method and system
CN110175575A (en) * 2019-05-29 2019-08-27 南京邮电大学 A kind of single Attitude estimation method based on novel high-resolution network model
CN110210402A (en) * 2019-06-03 2019-09-06 北京卡路里信息技术有限公司 Feature extracting method, device, terminal device and storage medium
CN110276316B (en) * 2019-06-26 2022-05-24 电子科技大学 Human body key point detection method based on deep learning
CN110276316A (en) * 2019-06-26 2019-09-24 电子科技大学 A kind of human body critical point detection method based on deep learning
CN110378253B (en) * 2019-07-01 2021-03-26 浙江大学 Real-time key point detection method based on lightweight neural network
CN110378253A (en) * 2019-07-01 2019-10-25 浙江大学 A kind of real time critical point detecting method based on lightweight neural network
CN110443144A (en) * 2019-07-09 2019-11-12 天津中科智能识别产业技术研究院有限公司 A kind of human body image key point Attitude estimation method
CN110349164A (en) * 2019-07-19 2019-10-18 北京华捷艾米科技有限公司 A kind of image, semantic dividing method, device and terminal device
CN111104841A (en) * 2019-09-16 2020-05-05 平安科技（深圳）有限公司 Violent behavior detection method and system
WO2021051547A1 (en) * 2019-09-16 2021-03-25 平安科技（深圳）有限公司 Violent behavior detection method and system
CN110969087B (en) * 2019-10-31 2023-11-21 杭州未名信科科技有限公司 Gait recognition method and system
CN110969087A (en) * 2019-10-31 2020-04-07 浙江省北大信息技术高等研究院 Gait recognition method and system
CN110929638B (en) * 2019-11-20 2023-03-07 北京奇艺世纪科技有限公司 Human body key point identification method and device and electronic equipment
CN110929638A (en) * 2019-11-20 2020-03-27 北京奇艺世纪科技有限公司 Human body key point identification method and device and electronic equipment
CN111062261A (en) * 2019-11-25 2020-04-24 维沃移动通信（杭州）有限公司 Image processing method and device
CN110889858A (en) * 2019-12-03 2020-03-17 中国太平洋保险（集团）股份有限公司 Automobile part segmentation method and device based on point regression
CN111291716A (en) * 2020-02-28 2020-06-16 深圳大学 Sperm cell recognition method, device, computer equipment and storage medium
CN111291716B (en) * 2020-02-28 2024-01-05 深圳市瑞图生物技术有限公司 Sperm cell identification method, sperm cell identification device, computer equipment and storage medium
CN111695519B (en) * 2020-06-12 2023-08-08 北京百度网讯科技有限公司 Method, device, equipment and storage medium for positioning key point
US11610389B2 (en) 2020-06-12 2023-03-21 Beijing Baidu Netcom Science And Technology Co., Ltd. Method and apparatus for positioning key point, device, and storage medium
CN111695519A (en) * 2020-06-12 2020-09-22 北京百度网讯科技有限公司 Key point positioning method, device, equipment and storage medium
CN112036244A (en) * 2020-07-30 2020-12-04 广东技术师范大学 Human body posture estimation method based on neural network
CN112131959B (en) * 2020-08-28 2024-03-22 浙江工业大学 2D human body posture estimation method based on multi-scale feature reinforcement
CN112131959A (en) * 2020-08-28 2020-12-25 浙江工业大学 2D human body posture estimation method based on multi-scale feature reinforcement
US11270147B1 (en) 2020-10-05 2022-03-08 International Business Machines Corporation Action-object recognition in cluttered video scenes using text
US11928849B2 (en) 2020-10-05 2024-03-12 International Business Machines Corporation Action-object recognition in cluttered video scenes using text
WO2022074483A1 (en) * 2020-10-05 2022-04-14 International Business Machines Corporation Action-object recognition in cluttered video scenes using text
GB2614170A (en) * 2020-10-05 2023-06-28 Ibm Action-object recognition in cluttered video scenes using text
GB2614170B (en) * 2020-10-05 2023-12-13 Ibm Action-object recognition in cluttered video scenes using text
TWI733616B (en) * 2020-11-04 2021-07-11 財團法人資訊工業策進會 Reconition system of human body posture, reconition method of human body posture, and non-transitory computer readable storage medium
CN112651294A (en) * 2020-11-05 2021-04-13 同济大学 Method for recognizing human body shielding posture based on multi-scale fusion
CN112418046B (en) * 2020-11-17 2023-06-23 武汉云极智能科技有限公司 Exercise guiding method, storage medium and system based on cloud robot
CN112418046A (en) * 2020-11-17 2021-02-26 武汉云极智能科技有限公司 Fitness guidance method, storage medium and system based on cloud robot
CN112597955B (en) * 2020-12-30 2023-06-02 华侨大学 Single-stage multi-person gesture estimation method based on feature pyramid network
CN112597955A (en) * 2020-12-30 2021-04-02 华侨大学 Single-stage multi-person attitude estimation method based on feature pyramid network
CN112966574A (en) * 2021-02-22 2021-06-15 厦门艾地运动科技有限公司 Human body three-dimensional key point prediction method and device and electronic equipment
CN113221626B (en) * 2021-03-04 2023-10-20 北京联合大学 Human body posture estimation method based on Non-local high-resolution network
CN113221626A (en) * 2021-03-04 2021-08-06 北京联合大学 Human body posture estimation method based on Non-local high-resolution network
US11423252B1 (en) 2021-04-29 2022-08-23 International Business Machines Corporation Object dataset creation or modification using labeled action-object videos
CN113033524A (en) * 2021-05-26 2021-06-25 北京的卢深视科技有限公司 Occlusion prediction model training method and device, electronic equipment and storage medium
CN117392761A (en) * 2023-12-13 2024-01-12 深圳须弥云图空间科技有限公司 Human body pose recognition method and device, electronic equipment and storage medium
CN117392761B (en) * 2023-12-13 2024-04-16 深圳须弥云图空间科技有限公司 Human body pose recognition method and device, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN108229445A (en) A kind of more people's Attitude estimation methods based on cascade pyramid network
CN109711316B (en) Pedestrian re-identification method, device, equipment and storage medium
Gao et al. Salient object detection in the distributed cloud-edge intelligent network
CN109886241A (en) Driver fatigue detection based on shot and long term memory network
CN110287960A (en) The detection recognition method of curve text in natural scene image
CN111695430B (en) Multi-scale face detection method based on feature fusion and visual receptive field network
Khosla et al. Looking beyond the visible scene
CN110135375A (en) More people's Attitude estimation methods based on global information integration
CN104680559B (en) The indoor pedestrian tracting method of various visual angles based on motor behavior pattern
CN110765833A (en) Crowd density estimation method based on deep learning
CN107480178A (en) A kind of pedestrian's recognition methods again compared based on image and video cross-module state
CN103006178B (en) Equipment based on three-dimensional motion following calculation energy expenditure and method
CN110414747A (en) A kind of space-time shot and long term urban human method for predicting based on deep learning
CN105740780A (en) Method and device for human face in-vivo detection
CN109034152A (en) License plate locating method and device based on LSTM-CNN built-up pattern
CN112767466B (en) Light field depth estimation method based on multi-mode information
US11106904B2 (en) Methods and systems for forecasting crowd dynamics
Li et al. Sign language recognition based on computer vision
CN109598225A (en) Sharp attention network, neural network and pedestrian's recognition methods again
KR20200040186A (en) Learning method and testing method for object detector based on r-cnn, and learning device and testing device using the same
Ling et al. Superresolution land cover mapping with multiscale information by fusing local smoothness prior and downscaled coarse fractions
Oliva et al. Representing, perceiving, and remembering the shape of visual space
CN110322509A (en) Object localization method, system and computer equipment based on level Class Activation figure
CN110647909A (en) Remote sensing image classification method based on three-dimensional dense convolution neural network
CN111178178B (en) Multi-scale pedestrian re-identification method, system, medium and terminal combined with region distribution

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
WW01 Invention patent application withdrawn after publication

Application publication date: 20180629

WW01 Invention patent application withdrawn after publication