CN109033946A - Merge the estimation method of human posture of directional diagram - Google Patents
Merge the estimation method of human posture of directional diagram Download PDFInfo
- Publication number
- CN109033946A CN109033946A CN201810588480.6A CN201810588480A CN109033946A CN 109033946 A CN109033946 A CN 109033946A CN 201810588480 A CN201810588480 A CN 201810588480A CN 109033946 A CN109033946 A CN 109033946A
- Authority
- CN
- China
- Prior art keywords
- thermal map
- network
- key point
- directional diagram
- feature
- 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
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
-
- 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
- G06T2207/10024—Color image
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention discloses a kind of estimation method of human posture for merging directional diagram, comprising: carries out feature extraction and obtains high-level feature, the primary learning detection and the detection of directional diagram for carrying out the key point location drawing obtain position thermal map and direction thermal map；The position thermal map, directional diagram and high-level feature of prediction are merged, more accurate position thermal map is obtained.The present invention improves key point precision of prediction, compared with previous network structure, the available more mankind's posture informations of the depth convolutional network proposed by fusion directional information.The present invention can effectively solve the problem that the Attitude estimation under occlusion issue, for the key point being blocked, when position, thermal map is in the main true, even if the location network of first stage cannot find key point position well, subsequent converged network can accurately also predict key point position, and robustness is more preferably.The network architecture parameters amount of this method is relatively much smaller, and in the training stage to the of less demanding of video memory, training is easier to restrain.
Description
Technical field
The invention belongs to computer visions, technical field of image processing, are related to a kind of human body attitude estimation technique, specifically
It says, is to be related to a kind of estimation method of human posture for merging directional diagram.
Background technique
It is universal with equipment such as smart phone, tablet computers, all the time for the image of physical activity, video data
It is all generating, how to allow computer automatic understanding human action also just to seem critically important, application field will be very extensive, such as people
Machine interaction, intelligent monitoring etc..Computer efficiently can automatically understand human action, and far-reaching shadow will be generated to entire society
It rings.In this background, human body attitude estimation is proposed out.The purpose of human body attitude estimation is exactly to detect the posture of human body
Behavior understands image, the human body attitude behavioural information in video by computer learning, obtains appearance that is digitized, can presenting
State information.Human body attitude estimation problem is a hot issue of computer vision field.Deep learning is machine learning research
In new field, the mechanism that it imitates human brain explains data, such as image, sound and text.With deep learning
Development, deep learning applied to multiple fields, and the contribution in computer vision field is especially prominent.Human body attitude is estimated
The research of meter aspect also obtains significant progress with the help of deep learning.
Up to the present, human body attitude estimation has been achieved for many research achievements, but all has such or such
Defect.Summing up can be divided into three categories: 1) using the structural model of visualization, these methods are both for human motion chain
Key position corresponding regional area is found in picture, optimize to obtain Attitude estimation model with this.But due to human body
Posture is flexible, and structural model is difficult to establish.2) position key point coordinate is returned using depth convolutional network.It is directly to make at the beginning
Key point coordinate is returned with convolutional network, and has cascaded multiple key points and has returned device, to improve the precision of recurrence.It
Afterwards, and occur returning the mode of key point thermal map using convolutional network.But the robustness of homing method is very poor, and model is expansible
Property is poor.3) method for detecting thermal map, i.e., place dimensional Gaussian thermal map at key point, and convolutional network study is allowed to generate thermal map,
Take the peak value of thermal map as the accurate coordinates of this key point.Such method has certain robustness, but for being blocked
Position key point or the position key point to reach unanimity with background color are difficult detected.
Carry out human body attitude estimation, need to solve many technical problems: (1) occlusion issue is most difficult to resolve in Attitude estimation
Certainly the problem of.It is divided into blocking certainly and is blocked with it, blocking certainly is exactly that human body itself has covered a part of key point, this is photograph taking
When angle problem caused by；It, which is blocked, refers to other objects blocking for human body key point.In the existing method, substantially
All it is direct detection position thermal map, allows relationship between the implicitly learned human body key position of depth convolutional neural networks, this causes
The human body blocked is difficult to detect, and has also slowed down the speed of training.(2) foreground picture is merged with Background, i.e. human body
Certain key point regions and the adjacent area color in Background are almost the same, this is resulted in can not be accurate in critical point detection
Key point position is estimated, causes detection error big.(3) it is difficult to distinguish between key point.For example, left and right foot is difficult in the case of leaning to one side
To distinguish.
Summary of the invention
To solve the above problems, the invention discloses it is a kind of merge directional diagram estimation method of human posture, be different from
Toward the method for directly detecting key point, the present invention allows convolutional network not only to learn key point location information, also study key point
Directional information.The present invention proposes directional diagram to solve the internal direction of key point, by a unified network architecture
In be integrated with key point position and the internal direction of physical feeling to consider the global posture information of human body, by depth convolution
Network phase merges the two information, completes the supervised learning to human body more information.
In order to achieve the above object, the invention provides the following technical scheme:
The Attitude estimation method for merging directional diagram, includes the following steps:
Step 1, it carries out feature extraction and obtains high-level feature, carry out primary learning detection and the side of the key point location drawing
Detection to figure obtains position thermal map and direction thermal map；
Step 2, the position thermal map of prediction, directional diagram and high-level feature are merged, obtains more accurate position
Set thermal map.
Further, the step 1 specifically includes as follows step by step:
Feature extraction: the RGB image given for a width, by its dimension normalization, by simple convolutional network, including
The convolutional layer of small convolution kernel, Relu activation primitive and batch normalizing operation, obtain basic image feature information；Su Yihou warp
A hourglass module is crossed, higher level feature is extracted；
Location network detection: it for the high-level feature extracted from hourglass network, is exercised supervision by location network
Learn position thermal map；
The detection of direction network: a selected key point as human body center, the direction thermal map of remaining key point according to its with
Dimensional Gaussian thermal map is placed in the direction at human body center, obtains direction thermal map.
Further, the step 2 specifically includes following process: first merging position thermal map with directional diagram, by convolution
Feature is obtained, then is added with the high-level feature that hourglass module extracts；Later, using one, only there are two dimensional informations
Hourglass module is handled, and final predicted position thermal map is obtained.
Compared with prior art, the invention has the advantages that and the utility model has the advantages that
By merging directional information, key point precision of prediction, compared with previous network structure, the depth that is proposed are improved
The available more mankind's posture informations of convolutional network.Therefore, the present invention can effectively solve the problem that the posture under occlusion issue is estimated
Meter, for the key point being blocked, when thermal map is in the main true when position, even if the location network of first stage can not be well
Key point position is found, subsequent converged network can accurately also predict key point position, and robustness is more preferably.Also, we
The network architecture parameters amount of method is relatively much smaller, and in the training stage to the of less demanding of video memory, training is easier to restrain.
Detailed description of the invention
Fig. 1 is depth convolutional network structural framing figure of the invention.
Fig. 2 is actual position thermal map (top half) and the true directions figure (lower half of part human body key point
Point).
Fig. 3 be the first stage predicted position thermal map (top) of part human body key point, prediction direction figure (middle part),
Second stage position thermal map (lower part).
Specific embodiment
Technical solution provided by the invention is described in detail below with reference to specific embodiment, it should be understood that following specific
Embodiment is only illustrative of the invention and is not intended to limit the scope of the invention.
The estimation method of human posture of fusion directional diagram provided by the invention is realized end and is arrived using depth convolutional network
The study at end, network structure are as shown in Figure 1.The RGB picture of input is first subjected to preliminary treatment using simple convolutional network, is obtained
To basic feature.Later, higher level feature is obtained by a hourglass network module.Again respectively by location network with
Direction network obtains position thermal map and direction thermal map.The converged network of the latter half is by position thermal map, directional diagram and before
High-level feature merged, obtain detecting more accurate position thermal map.
Specifically, including two stages.First stage carries out primary learning detection and the direction of the key point location drawing
The detection of figure, second stage are merged, and more accurate position thermal map is obtained.Include multiple supervised learning processes.
The depth convolutional network of first stage mainly includes: feature extraction network, location network and direction network.
Feature extraction: its dimension normalization is 256*256, by simple convolution by the RGB image given for a width
Network, the convolutional layer including small convolution kernel, Relu activation primitive and batch normalizing operation obtain basic characteristics of image letter
Breath, the dimension of feature is 64*64*256 at this time.Later, by a hourglass module, it can extract higher level feature.Hourglass
Network module passes through multiple down-sampling and up-sampling, combines the feature under multiple scales, can more comprehensively describe to scheme
As feature.In the present invention, we combine the feature under 4 scales, i.e. 8*8,16*16,32*32 and 64*64, can incite somebody to action
The information of human body is comprehensively included.The dimension of feature is 64*64*256 at this time.
Location network: it for the high-level feature extracted from hourglass network, is exercised supervision study by location network
Position thermal map illustrates Partial key point as shown in top half in Fig. 2, in figure.
Location network is made of a residual error module, a full articulamentum, a convolutional layer.Last convolutional layer exports J
A position thermal map, scale are J*64*64 (J is key point number).Loss function herein are as follows:
In above formula, p indicates p-th of position position,The prediction result of location network is characterized,Represent position heat
The legitimate reading of figure.
Direction network: it is similar with location network, and by a residual error module, a full articulamentum, a convolutional layer structure
At.The high-level feature that hourglass network is extracted is as input, and exercise supervision study.
Similar to position thermal map, the generation of directional information is also by a dimensional Gaussian, i.e. direction thermal map.Initialization one
It is a be all 0 two dimensional character figure, calculate the direction relations of key point and human body center, it is hot to place dimensional Gaussian according to this relationship
Figure, peak value be 1 to get arrive direction thermal map.We set navel as human body center, i.e., the directional diagram of navel is Gauss heat
Figure is in intermediate two dimensional character figure.The directional diagram of remaining key point is related to navel.For example, right crus of diaphragm ankle is under navel
Side, then just in the lower section of white circle, (note: white circle is intended merely to visual observation and conveniently draws the directional diagram of right crus of diaphragm ankle
System).Lower half portion illustrates the directional diagram of Partial key point in Fig. 2.
Finally export J directional diagram, scale J*64*64.Loss function herein are as follows:
In above formula,The prediction result of direction network is characterized,Represent the legitimate reading of directional diagram.
The network of second stage is converged network, and the position thermal map, directional diagram and hourglass network of prediction are extracted
High-level feature merged.Position thermal map is merged with directional diagram first, obtains the feature of 64*64*256 by convolution, then
It is added with the high-level feature that hourglass module extracts.Later, using one only there are two dimensional information hourglass module into
Row processing, obtains final predicted position thermal map.Supervised learning target and the location network of first stage herein supervises target
Unanimously, loss function are as follows:
The prediction result of converged network is characterized,The legitimate reading for representing position thermal map, withUnanimously.
Complete loss function are as follows:
L=L1+L2+L3
Training is realized
80 wheel training are carried out on MPII data set.When training, data enhancing has been carried out to training data.All images contracting
It is put into 256*256, Random-Rotation ± 30 °, stochastic transformation size (0.75-1.25), and (0.8-1.2) is enhanced to RGB color.This
So that our model has more robustness.It is realized using PyTorch, using RMSprop optimization algorithm, initial study is set
Rate is 5 × 10-4。
It after training, is tested on MPII test set, has obtained goodr result.It is visually dissolved
Come it was found that maximum promotion effect is that the prediction for sheltering part is more accurate, such as Fig. 3.Fig. 3 (on) it is first
For the Partial key point prediction that stage position network obtains as a result, it can be found that there is the right crus of diaphragm ankle blocked, prediction is inadequate
Accurately.And the directional diagram that direction network obtains be accurately, as Fig. 3 (in).The position thermal map obtained after converged network
Can obtain accurate prediction result, as Fig. 3 (under).
The technical means disclosed in the embodiments of the present invention is not limited only to technological means disclosed in above embodiment, further includes
Technical solution consisting of any combination of the above technical features.It should be pointed out that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (3)
1. merging the Attitude estimation method of directional diagram, which comprises the steps of:
Step 1, it carries out feature extraction and obtains high-level feature, carry out primary learning detection and the directional diagram of the key point location drawing
Detection obtain position thermal map and direction thermal map；
Step 2, the position thermal map of prediction, directional diagram and high-level feature are merged, obtains more accurate position heat
Figure.
2. the Attitude estimation method of fusion directional diagram according to claim 1, which is characterized in that the step 1 is specifically wrapped
It includes as follows step by step:
Feature extraction: the RGB image given for a width, by its dimension normalization, by simple convolutional network, including rouleau
The convolutional layer of product core, Relu activation primitive and batch normalizing operation, obtain basic image feature information；Su Yihou passes through one
A hourglass module, extracts higher level feature；
Location network detection: it for the high-level feature extracted from hourglass network, is exercised supervision study by location network
Position thermal map；
The detection of direction network: a key point is selected as human body center, the direction thermal map of remaining key point is according to itself and human body
Dimensional Gaussian thermal map is placed in the direction at center, obtains direction thermal map.
3. the Attitude estimation method of fusion directional diagram according to claim 1, which is characterized in that the step 2 is specifically wrapped
It includes following process: first merging position thermal map with directional diagram, obtain feature, then the height extracted with hourglass module by convolution
Level characteristics are added；Later, using one, only there are two the hourglass modules of dimensional information to be handled, and obtains final prediction bits
Set thermal map.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810588480.6A CN109033946A (en) | 2018-06-08 | 2018-06-08 | Merge the estimation method of human posture of directional diagram |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810588480.6A CN109033946A (en) | 2018-06-08 | 2018-06-08 | Merge the estimation method of human posture of directional diagram |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109033946A true CN109033946A (en) | 2018-12-18 |
Family
ID=64612640
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810588480.6A Pending CN109033946A (en) | 2018-06-08 | 2018-06-08 | Merge the estimation method of human posture of directional diagram |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109033946A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685023A (en) * | 2018-12-27 | 2019-04-26 | 深圳开立生物医疗科技股份有限公司 | A kind of facial critical point detection method and relevant apparatus of ultrasound image |
| CN109858333A (en) * | 2018-12-20 | 2019-06-07 | 腾讯科技（深圳）有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN109934111A (en) * | 2019-02-12 | 2019-06-25 | 清华大学深圳研究生院 | A kind of body-building Attitude estimation method and system based on key point |
| CN110084161A (en) * | 2019-04-17 | 2019-08-02 | 中山大学 | A kind of rapid detection method and system of skeleton key point |
| 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 |
| 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 |
| CN110490080A (en) * | 2019-07-22 | 2019-11-22 | 西安理工大学 | A kind of human body tumble method of discrimination based on image |
| CN110569724A (en) * | 2019-08-05 | 2019-12-13 | 湖北工业大学 | Face alignment method based on residual hourglass network |
| CN110610154A (en) * | 2019-09-10 | 2019-12-24 | 北京迈格威科技有限公司 | Behavior recognition method and apparatus, computer device, and storage medium |
| CN110969105A (en) * | 2019-11-22 | 2020-04-07 | 清华大学深圳国际研究生院 | Human body posture estimation method |
| CN111191622A (en) * | 2020-01-03 | 2020-05-22 | 华南师范大学 | Posture recognition method and system based on thermodynamic diagram and offset vector and storage medium |
| CN111724901A (en) * | 2020-06-19 | 2020-09-29 | 中山大学 | Method, system and device for predicting structure body parameters based on vision and storage medium |
| CN111797791A (en) * | 2018-12-25 | 2020-10-20 | 上海智臻智能网络科技股份有限公司 | Human body posture recognition method and device |
| CN112102223A (en) * | 2019-06-18 | 2020-12-18 | 通用电气精准医疗有限责任公司 | Method and system for automatically setting scanning range |
| CN112364738A (en) * | 2020-10-30 | 2021-02-12 | 深圳点猫科技有限公司 | Human body posture estimation method, device, system and medium based on deep learning |
| CN113408568A (en) * | 2021-04-16 | 2021-09-17 | 科大讯飞股份有限公司 | Method, device and equipment for correlating detection model training of object key points |
| CN116385829A (en) * | 2023-04-07 | 2023-07-04 | 北京百度网讯科技有限公司 | Gesture description information generation method, model training method and device |
| CN112102223B (en) * | 2019-06-18 | 2024-05-14 | 通用电气精准医疗有限责任公司 | Method and system for automatically setting scan range |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650827A (en) * | 2016-12-30 | 2017-05-10 | 南京大学 | Human body posture estimation method and system based on structure guidance deep learning |
| CN106991386A (en) * | 2017-02-27 | 2017-07-28 | 杭州电子科技大学 | A kind of gesture identification method based on depth residual error network |
-
2018
- 2018-06-08 CN CN201810588480.6A patent/CN109033946A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650827A (en) * | 2016-12-30 | 2017-05-10 | 南京大学 | Human body posture estimation method and system based on structure guidance deep learning |
| CN106991386A (en) * | 2017-02-27 | 2017-07-28 | 杭州电子科技大学 | A kind of gesture identification method based on depth residual error network |
Non-Patent Citations (2)
| Title |
|---|
| WENLIN ZHUANG: "《Human Pose Estimation using DirectionMaps》", 《THE 33RD YOUTH ACADEMIC ANNUAL CONFERENCE OF CHINESE ASSOCIATION OF AUTOMATION》 * |
| 魏小峰 等；: "《空间目标三维姿态估计方法》", 《武汉大学学报· 信息科学版》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109858333B (en) * | 2018-12-20 | 2023-01-17 | 腾讯科技（深圳）有限公司 | Image processing method, image processing device, electronic equipment and computer readable medium |
| CN109858333A (en) * | 2018-12-20 | 2019-06-07 | 腾讯科技（深圳）有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN111797791A (en) * | 2018-12-25 | 2020-10-20 | 上海智臻智能网络科技股份有限公司 | Human body posture recognition method and device |
| CN109685023A (en) * | 2018-12-27 | 2019-04-26 | 深圳开立生物医疗科技股份有限公司 | A kind of facial critical point detection method and relevant apparatus of ultrasound image |
| CN109934111B (en) * | 2019-02-12 | 2020-11-24 | 清华大学深圳研究生院 | Fitness posture estimation method and system based on key points |
| CN109934111A (en) * | 2019-02-12 | 2019-06-25 | 清华大学深圳研究生院 | A kind of body-building Attitude estimation method and system based on key point |
| CN110084161A (en) * | 2019-04-17 | 2019-08-02 | 中山大学 | A kind of rapid detection method and system of skeleton key point |
| 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 |
| 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 |
| CN112102223B (en) * | 2019-06-18 | 2024-05-14 | 通用电气精准医疗有限责任公司 | Method and system for automatically setting scan range |
| CN112102223A (en) * | 2019-06-18 | 2020-12-18 | 通用电气精准医疗有限责任公司 | Method and system for automatically setting scanning range |
| CN110490080A (en) * | 2019-07-22 | 2019-11-22 | 西安理工大学 | A kind of human body tumble method of discrimination based on image |
| CN110569724B (en) * | 2019-08-05 | 2021-06-04 | 湖北工业大学 | Face alignment method based on residual hourglass network |
| CN110569724A (en) * | 2019-08-05 | 2019-12-13 | 湖北工业大学 | Face alignment method based on residual hourglass network |
| CN110610154A (en) * | 2019-09-10 | 2019-12-24 | 北京迈格威科技有限公司 | Behavior recognition method and apparatus, computer device, and storage medium |
| CN110969105A (en) * | 2019-11-22 | 2020-04-07 | 清华大学深圳国际研究生院 | Human body posture estimation method |
| CN111191622A (en) * | 2020-01-03 | 2020-05-22 | 华南师范大学 | Posture recognition method and system based on thermodynamic diagram and offset vector and storage medium |
| CN111191622B (en) * | 2020-01-03 | 2023-05-26 | 华南师范大学 | Gesture recognition method, system and storage medium based on thermodynamic diagram and offset vector |
| CN111724901A (en) * | 2020-06-19 | 2020-09-29 | 中山大学 | Method, system and device for predicting structure body parameters based on vision and storage medium |
| CN112364738A (en) * | 2020-10-30 | 2021-02-12 | 深圳点猫科技有限公司 | Human body posture estimation method, device, system and medium based on deep learning |
| CN113408568A (en) * | 2021-04-16 | 2021-09-17 | 科大讯飞股份有限公司 | Method, device and equipment for correlating detection model training of object key points |
| CN113408568B (en) * | 2021-04-16 | 2024-04-16 | 科大讯飞股份有限公司 | Related method, device and equipment for training detection model of object key points |
| CN116385829A (en) * | 2023-04-07 | 2023-07-04 | 北京百度网讯科技有限公司 | Gesture description information generation method, model training method and device |
| CN116385829B (en) * | 2023-04-07 | 2024-02-06 | 北京百度网讯科技有限公司 | Gesture description information generation method, model training method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109033946A (en) | Merge the estimation method of human posture of directional diagram | |
| US11908244B2 (en) | Human posture detection utilizing posture reference maps | |
| CN110348330B (en) | Face pose virtual view generation method based on VAE-ACGAN | |
| Tsai et al. | Implementation of fall detection system based on 3D skeleton for deep learning technique | |
| CN103218605B (en) | A kind of fast human-eye positioning method based on integral projection and rim detection | |
| CN109902548B (en) | Object attribute identification method and device, computing equipment and system | |
| CN108875533B (en) | Face recognition method, device, system and computer storage medium | |
| CN112733767B (en) | Human body key point detection method and device, storage medium and terminal equipment | |
| Xia et al. | Human motion recovery jointly utilizing statistical and kinematic information | |
| CN113033369B (en) | Motion capture method, motion capture device, electronic equipment and computer readable storage medium | |
| WO2023151237A1 (en) | Face pose estimation method and apparatus, electronic device, and storage medium | |
| CN113111767A (en) | Fall detection method based on deep learning 3D posture assessment | |
| CN110222780A (en) | Object detecting method, device, equipment and storage medium | |
| CN112287865B (en) | Human body posture recognition method and device | |
| CN115375737B (en) | Target tracking method and system based on adaptive time and serialized space-time characteristics | |
| CN111507184B (en) | Human body posture detection method based on parallel cavity convolution and body structure constraint | |
| Yin et al. | Estimation of the fundamental matrix from uncalibrated stereo hand images for 3D hand gesture recognition | |
| CN112699857A (en) | Living body verification method and device based on human face posture and electronic equipment | |
| CN110728172B (en) | Point cloud-based face key point detection method, device and system and storage medium | |
| CN116758212A (en) | 3D reconstruction method, device, equipment and medium based on self-adaptive denoising algorithm | |
| CN110705355A (en) | Face pose estimation method based on key point constraint | |
| CN106662927A (en) | Motion recognition method and motion recognition device | |
| CN107609479A (en) | Attitude estimation method and system based on the sparse Gaussian process with noise inputs | |
| Bhuyan et al. | Hand gesture recognition and animation for local hand motions | |
| CN113658180A (en) | Surface defect region segmentation method and device based on spatial context guidance |
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: 20181218 |