CN109117753A - Position recognition methods, device, terminal and storage medium - Google Patents
Position recognition methods, device, terminal and storage medium Download PDFInfo
- Publication number
- CN109117753A CN109117753A CN201810820840.0A CN201810820840A CN109117753A CN 109117753 A CN109117753 A CN 109117753A CN 201810820840 A CN201810820840 A CN 201810820840A CN 109117753 A CN109117753 A CN 109117753A
- Authority
- CN
- China
- Prior art keywords
- module
- user
- identification model
- submodel
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/442—Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
- H04N21/44213—Monitoring of end-user related data
- H04N21/44218—Detecting physical presence or behaviour of the user, e.g. using sensors to detect if the user is leaving the room or changes his face expression during a TV program
-
- 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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Social Psychology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Databases & Information Systems (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of position recognition methods, device, terminal and storage mediums.Position recognition methods includes: the pose presentation for obtaining at least one position for showing user；The pose presentation is input in the identification model of position；Obtain the location information at least one position of the user of the position identification model output；Wherein, the position identification model includes feature extraction submodel and location detection submodel, and the feature extraction submodel includes: the sequentially connected convolution module of multiple groups and down-sampled module；The location detection submodel includes: multiple sequentially connected detection modules and convolution module；The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling module.Recognition methods operand in position provided in an embodiment of the present invention is small, and structure is simple, meets the requirement of real-time of live streaming, can be in terminal operating.
Description
Technical field
The present embodiments relate to computer vision technique more particularly to a kind of position recognition methods, device, terminal and deposit
Storage media.
Background technique
With the development of direct seeding technique, more and more users obtain information, participation activity by live video, such as logical
It crosses live video and carries out posture correction or motion guiding.
More important link is to identify the position of user during carrying out posture correction or motion guiding.User's
Each position includes the organs such as each limbs, joint and eye ear mouth and nose.With the development of computer vision technique, image can be passed through
Recognizer identifies the information such as position, the size at each position in image, provides posture correction or motion guiding suggestion for user.
But structure is complicated, operand is big for existing image recognition algorithm, needs biggish memory space and longer fortune
Evaluation time causes to be difficult in terminal operating, and is difficult to meet the requirement of real-time of live streaming.
Summary of the invention
The embodiment of the present invention provides a kind of position recognition methods, device, terminal and storage medium, to realize user position
Quickly identification, meets the requirement of real-time of live streaming, can be in terminal operating.
In a first aspect, the embodiment of the invention provides a kind of position recognition methods, comprising:
Obtain the pose presentation for showing at least one position of user；
The pose presentation is input in the identification model of position；
Obtain the location information at least one position of the user of the position identification model output；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction
Submodel includes: that the sequentially connected convolution module of multiple groups and down-sampled module for inputting pose presentation export the posture figure
Feature as in；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, described in inputting
Feature in pose presentation exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling mould
Block.
Second aspect, the embodiment of the invention also provides a kind of position identification device, which includes:
First obtains module, for obtaining the pose presentation for showing at least one position of user；
Input module, for the pose presentation to be input in the identification model of position；
Second obtains module, the position letter at least one position of the user for obtaining the position identification model output
Breath；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction
Submodel includes: that the sequentially connected convolution module of multiple groups and down-sampled module for inputting pose presentation export the posture figure
Feature as in；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, described in inputting
Feature in pose presentation exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling mould
Block.
The third aspect, the embodiment of the invention also provides a kind of terminals, comprising:
One or more processors；
Memory, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes position recognition methods described in any embodiment.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer
Program realizes position recognition methods described in any embodiment when the program is executed by processor.
The embodiment of the invention provides a kind of position identification models, realize a kind of position recognizer based on this model.
Wherein, position identification model includes the feature extraction submodel for feature extraction and the location detection submodule for position identification
Type, feature extraction submodel include the sequentially connected convolution module of multiple groups and down-sampled module, by by sampling module to extraction
To feature zoom in and out, to reduce intrinsic dimensionality, reduce data redundancy；Location detection submodel includes: multiple successively to connect
The detection module and convolution module connect is gradually corrected recognition result, by multiple connected detection modules to reach
Preferable recognition effect；Detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling
Module, rise sampling is converted to high-definition picture to the low-resolution image received, differentiates enough to get to have
The location information of rate, it is seen then that the structure of position identification model is simple, parameter is few, can be stored in terminal and recognition speed is very fast,
And then the position recognizer based on position identification model can satisfy the requirement of real-time of live streaming in terminal operating.
Detailed description of the invention
Fig. 1 a is a kind of flow chart for position recognition methods that the embodiment of the present invention one provides；
Fig. 1 b is a kind of structural schematic diagram for position identification model that the embodiment of the present invention one provides；
Fig. 2 is a kind of structural schematic diagram of position identification model provided by Embodiment 2 of the present invention；
Fig. 3 a is a kind of flow chart for position recognition methods that the embodiment of the present invention three provides；
Fig. 3 b is the thermodynamic chart that the embodiment of the present invention three provides；
Fig. 3 c is the posture information schematic diagram for the user that the embodiment of the present invention three provides；
Fig. 4 is a kind of structural schematic diagram for position identification device that the embodiment of the present invention six provides；
Fig. 5 is a kind of structural schematic diagram for terminal that the embodiment of the present invention seven provides.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Embodiment one
Fig. 1 a is a kind of flow chart for position recognition methods that the embodiment of the present invention one provides, and the present embodiment is applicable to
Terminal acquire user pose presentation, and according to pose presentation identify user position the case where, this method can be known by position
Other device executes, which can be by hardware and/or software sharing, and is integrated in terminal, specifically comprise the following steps:
S110, acquisition show the pose presentation at least one position of user.
Optionally, pose presentation can be acquires in real time, is also possible to pre-stored.
The mode acquired in real time is specifically described below: terminal is equipped with the camera towards user side, passes through camera
User is shot, to acquire the pose presentation of user.It is alternatively possible in preset time acquisition pose presentation, take pictures
Pose presentation or periodically acquisition pose presentation are acquired after instruction.Pose presentation can be one, two or multiple.
The user position shown on pose presentation includes but is not limited to head, neck, left/right shoulder, left/right elbow, left/right
Wrist, left/right stern, left/right knee, left/right ankle.In some embodiments, user position can be described as the key point of user.
S120, pose presentation is input in the identification model of position.
The location information at least one position for the user that S130, acquisition position identification model export.
The input of position identification model is pose presentation, and output is the location information at least one position in pose presentation.
Based on this, pose presentation is input to position identification model, and obtains the position at least one position of position identification model output
Confidence breath.Optionally, the location information at least one position can be position coordinates or regional scope, such as thermodynamic chart.
Fig. 1 b is a kind of structural schematic diagram for position identification model that the embodiment of the present invention one provides.As shown in Figure 1 b, it wraps
Include: feature extraction submodel 11 and location detection submodel 12, the feature extraction submodel 11 include: that multiple groups are sequentially connected
Convolution module 13 and down-sampled module 14 export the feature in the pose presentation for inputting pose presentation.
Convolution module 13 is used to carry out feature extraction to the content of input.Convolution module 13 includes at least one for extracting
The convolutional layer of feature.The feature progress that down-sampled module 14 is used to export convolution module 13 is down-sampled, optionally, using max
It is down-sampled that pooling method carries out 1/2 to the feature that convolution module 13 exports.Down-sampled purpose be to the feature received into
Row scaling reduces data redundancy to reduce intrinsic dimensionality.
Location detection submodel 12 include: multiple sequentially connected detection modules 15 (as shown in dotted line frame in Fig. 1 b) and
Convolution module 13, the feature for inputting in the pose presentation, exports the location information of corresponding position.
Wherein, detection module 15 includes: the sequentially connected convolution module 13 of down-sampled module 14, multiple groups being connected and rises
Sampling module 16.By multiple connected detection modules 15, recognition result is gradually corrected, to reach preferable identification
Effect.2 detection modules 15 of sequential connection are shown in Fig. 1 b, but not limited to this, it can be in location detection submodel 12
Multiple detection modules 15 including sequential connection, it is demonstrated experimentally that the recognition effect reached is relatively managed when using 5 detection modules 15
Think.
It rises sampling module 16 and is used to carry out the feature that convolution module 13 exports a liter sampling, it is optionally, poor using arest neighbors
Value or bilinearity difference arithmetic carry out 1/2 liter of sampling to the feature that convolution module 13 exports.The purpose for rising sampling is to reception
To low-resolution image be converted to high-definition picture, to get the image with enough resolution ratio.
The embodiment of the invention provides a kind of position identification models, realize a kind of position recognizer based on this model.
Wherein, position identification model includes the feature extraction submodel for feature extraction and the location detection submodule for position identification
Type, feature extraction submodel include the sequentially connected convolution module of multiple groups and down-sampled module, by by sampling module to extraction
To feature zoom in and out, to reduce intrinsic dimensionality, reduce data redundancy；Location detection submodel includes: multiple successively to connect
The detection module and convolution module connect is gradually corrected recognition result, by multiple connected detection modules to reach
Preferable recognition effect；Detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling
Module, rise sampling is converted to high-definition picture to the low-resolution image received, differentiates enough to get to have
The location information of rate, it is seen then that simply, correspondingly, parameter is less, can be stored in terminal and identification for the structure of position identification model
Fast speed, and then the position recognizer based on position identification model can satisfy the real-time of live streaming in terminal operating
Property require.
Embodiment two
The present embodiment advanced optimizes the position identification model in above-described embodiment, specifically in position identification model
Middle increase skip floor.Fig. 2 is a kind of structural schematic diagram of position identification model provided by Embodiment 2 of the present invention, on the basis of Fig. 1 b
On, which further includes skip floor 17.
By many experiments, if position identification model includes feature extraction submodel and location detection submodel, identification
As a result not ideal enough, such as there is the situation of position identification mistake.Based on this, in feature extraction submodel and location detection submodule
Between type, increase skip floor 17.Skip floor 17 is used for the spy for exporting the down-sampled module 14 of at least one in feature extraction submodel 11
After sign is spliced, the feature for rising the output of sampling module 16 corresponding with location detection submodel 12 is spliced.
Wherein, skip floor 17 splices the feature that down-sampled modules 14 different in feature extraction submodel 11 export；So
Afterwards, by aforementioned spliced feature respectively with the spy of equal resolution that rises sampling module 16 in location detection submodel 12 and export
Sign is spliced.It is worth noting that the resolution ratio for the feature mutually spliced answer it is identical.If resolution ratio is different, liter can be used
Resolution adjustment is identical by sampling module 16 or down-sampled module 14.3 are shown in Fig. 2 for adjusting resolution ratio
Module, the fill color of the module for adjusting resolution ratio are black.
In the present embodiment, by increasing skip floor 17, study is arrived multiple dimensioned, global characteristics by location detection submodel 12,
To which location detection submodel 22 is able to get enough global informations, and then position identification model 22 can be from complicated appearance
The location information of corresponding position is recognized accurately in state image.
In some embodiments, each convolution module 13 includes M bottleneck18, and each bottleneck18 includes
Expansion layer, convolutional layer and the compression layer being connected.This is because the feature that convolutional layer extracts is limited to the port number of input, such as
First " compression " channel of fruit, then feature is extracted, then extractible feature is less.Therefore, first expanding channel, then extracted by convolution
Feature, then pressure channel, to extract sufficient amount of feature.In MobileNet-V2 network structure
Bottleneck is identical, and in the present embodiment, expansion layer includes the convolution kernel of 1*1, and convolutional layer includes the convolution kernel of 3*3, compression layer
Convolution kernel including 1*1.Optionally, through lot of experiment validation, when 3≤M≤5, identification accuracy is higher.
Output channel number in Bottleneck18 is N, 16≤N≤96, than defeated in MobileNet-V2 network structure
Port number is few out, to reduce network parameter, improves computational efficiency；Space needed for saving storage network simultaneously, thus
Enable position identification model in terminal operating.
(i.e. bottleneck is a for liter channel multiple, convolution size, output channel number and the number of repetition of Bottleneck18
Number) as shown in Fig. 2, being marked in the top of corresponding convolution module.Certainly, the parameter of the convolution module 13 in Fig. 2 is only example, this
Field technical staff can be adjusted each parameter of convolution module, to obtain preferable recognition effect.
It is worth noting that under normal circumstances, the convolution module in feature extraction submodel 11, location detection submodel 12
More, the bottleneck that each convolution module includes is more, then the feature extracted is deeper, and the recognition effect reached is better.But
It is that will lead to effect variation when network reaches certain depth.Those skilled in the art can Fig. 2 shows network knot
On the basis of structure, the bottleneck quantity that the quantity of detection module, the quantity of convolution module, each convolution module include is adjusted,
To realize preferable recognition effect.
In the present embodiment, at least one bottleneck constitutes convolution module and carries out feature extraction, can extract multi-pass
Road, sufficient amount of feature；The output channel number in output channel number ratio MobileNet-V2 network structure in convolution module
It is few, to reduce network parameter, improve computational efficiency；Space needed for saving storage network simultaneously, so as at end
End carries out image recognition；It is down-sampled by the feature progress exported to convolution module, i.e., feature is zoomed in and out, to reduce spy
Dimension is levied, data redundancy is reduced；Make the study of location detection submodel to multiple dimensioned, global characteristics by skip floor, thus position
Detection submodel is able to get enough global informations, and then position identification model can be accurate from complicated pose presentation
Identify the location information of corresponding position.
In some embodiments, in order to improve location detection model in the operating rate of terminal, with separable convolution module
Instead of bottleneck.
Specific proof procedure is as follows:
Assuming that the input of convolution is having a size of (DF,DF, M), the convolution kernel K used is (DK,DK, M, N), step-length 1, output
Characteristic pattern G size be (DG,DG,N).The calculation amount of so traditional convolution are as follows: DK·DK·M·N·DF·DF, wherein M be
Input channel number, N are output channel number.
Separable convolution is then that conventional roll has been integrated into the point-by-point convolution of depth convolution sum, corresponding calculation amount are as follows: DK·
DK·M·DF·DF+M·N·DF·DF, the ratio of two kinds of convolutional calculation amounts is as follows:
It can be seen from ratio under same hyper parameter, the calculation amount for separating convolution is traditional convolution calculation amountTimes.Moreover, also it is not much different in precision, in the case of same model and input data, traditional convolution
Probability of correct keypoint (PCKh)=98, and PCKh=97.2 under separable convolution)
Embodiment three
Fig. 3 a is a kind of flow chart for position recognition methods that the embodiment of the present invention three provides, comprising the following steps:
S310, acquisition show the pose presentation at least one position of user.
S320, pose presentation is input in the identification model of position.
S330, the thermodynamic chart for obtaining user's corresponding position that position identification model exports.
In the present embodiment, which is input to position identification model, the corresponding each position of position identification model output
Thermodynamic chart, a thermodynamic chart shows a position.Certainly, a thermodynamic chart can be with two positions or multiple positions.This
Thermodynamic chart in embodiment is the diagram that the region where corresponding position is shown in special highlighted form.
As shown in Figure 3b, it is assumed that position identification model exports 3 thermodynamic charts altogether, first be corresponding left wrist thermodynamic chart,
Second be corresponding right wrist thermodynamic chart, third is the thermodynamic chart of corresponding neck.
S340, from the thermodynamic chart at each position of correspondence, determine the position coordinates at each position respectively.
Optionally, from the highlight regions on the thermodynamic chart at each position optionally a bit (x, y), the position as corresponding position
Coordinate.Preferably, central point or most bright point are selected from the highlight regions on the thermodynamic chart at each position, as corresponding position
Position coordinates.Specifically, the thermodynamic chart at each position is filtered (such as gaussian filtering, Wiener filtering or mean filter
Deng), screening gray scale is minimum in image after the filtering or position coordinates of the coordinate of the highest point of brightness as corresponding position.
S350, the position coordinates according at least one position, calculate the posture information of user.
When user body parts are in different positions, the posture that user is shown is also different, such as left wrist is located at the right side
On wrist, it is located on left wrist with right wrist, shows the different posture of user.In the present embodiment, the posture information of user includes
Relative angle of angle, the orientation of limbs etc. of at least one position composition between information and/or limbs and limbs, it is opposite away from
From etc..Certainly, different according to the type of exercise of user, the requirement to posture is also different, any to be believed by the position at position
Cease obtained posture information all within the scope of protection of this application.
Optionally, the location information at least one position can be the thermodynamic chart of corresponding position or determine from thermodynamic chart
Each position position coordinates.
It is worth noting that if location information is position coordinates, directly according to the location information at least one position, meter
Calculate the posture information of user.Specifically, according to the location information at least one position, the bending between the connected limbs of user is calculated
Angle；And/or the location information according at least one position, calculate orientation angles of each limbs of user relative to preset direction.
If location information is thermodynamic chart, S340 is first needed to be implemented according to the location information at least one position, determines corresponding position
After coordinate, further according to the position coordinates at least one position, the posture information of user is calculated.Specifically, according at least one portion
The position coordinates of position calculate the bending angle between the connected limbs of user；And/or the position coordinates according at least one position,
Calculate orientation angles of each limbs of user relative to preset direction.
Connected limbs are, for example, upper arm and lower arm, H&N, thigh and shank etc..Be connected limbs between bending angle be
Angle between connected two limbs.Preset direction can be horizontally oriented and/or vertical direction.
Fig. 3 c is the posture information schematic diagram for the user that the embodiment of the present invention three provides, and the position that Fig. 3 c is shown includes a left side
Shoulder, left elbow and left wrist.Upper arm vectorIt indicates,Lower arm vectorIt indicates,Bending angle between upper arm and lower arm
The orientation angles calculation method of upper arm is as follows: assuming that preset direction is vertically oriented, corresponding vector
Calculate vector-To vectorAngle h.
S360, posture information and reference attitude information according to user prompt the user with posture correction information.
Reference attitude information corresponding with the posture information of user is stored in advance, such as the posture information of user is tree formula,
Then reference attitude information is the tree formula posture information of benchmark.
By comparing the posture information and reference attitude information of user, posture correction information is obtained, and posture is corrected and is believed
Breath is shown on terminal display screen and/or by voice prompting, to realize automatic, the timely correction of posture.
Illustratively, reference attitude information includes: that benchmark bending angle between connected limbs and/or each limbs are opposite
In the reference bearing angle of preset direction.
In the present embodiment, according to the posture information of user and reference attitude information, posture correction information, packet are prompted the user with
Include following two embodiment.
The first embodiment: firstly, according to bending angle and corresponding benchmark bending angle model between connected limbs
It encloses, calculates the first score of the limbs that are connected.Wherein, references angle range is by 3 threshold values: std, min, max composition.Std is indicated
Median, min indicate minimum value, and max indicates maximum value.Optionally, the first of connected limbs is calculated according to following piecewise functions
Score.Certainly, however it is not limited to this, it can also be directly using the difference of bending angle and std as the first score.
Wherein, x indicates the bending angle between connected limbs.
Then, the orientation angles and corresponding reference bearing angle range according to each limbs relative to preset direction calculate
Second score of each limbs.
Optionally, identical as the calculation method of the first score, reference bearing angle range is also by 3 threshold values: std, min,
Max composition.The second score of each limbs is calculated according to above-mentioned piecewise function.At this point, the x in piecewise function indicates the side of each limbs
Parallactic angle degree.
Then, summation is weighted to the first score and the second score, obtains the comprehensive score of user；Prompt the user with
The corresponding posture of comprehensive score corrects information.
Pass through formulaCalculate the comprehensive score of user.Wherein, n is the azimuth of each limbs
The sum of the quantity of bending angle between the quantity of degree and the limbs that are connected, xiBe i-th of limbs orientation angles or connected limb
Body orientation angles, f (xi) be i-th of limbs or connected limbs score, such as f (x1) indicate upper arm orientation angles
Point, f (x2) indicate the score of bending angle between upper arm and lower arm.wiIt is corresponding f (xi) weight,
The weight of weight and the second score of each limbs of the first score of limbs of being respectively connected can identical not yet
Together.Requirement in view of different postures to different limbs is different, assigns the first weight to the score of demanding limbs, to requiring
The score of low limbs assigns the second weight.First weight is greater than the second weight.
Different comprehensive scores corresponds to different posture correction information, and it is to user's body dimension which, which corrects information,
Correction.For example, the comprehensive score of user greater than the first fragmentation value, such as 80 points, then prompts, " user's flexibility is preferable, and four limbs relax
Exhibition, has integrally reached higher level "；The comprehensive score of user is greater than the second fragmentation value, such as 60 points, less than the first fragmentation value,
Then prompt " user's flexibility is normal, and four limbs are unfolded, and has integrally reached standard level "；The comprehensive score of user is less than the second segmentation
Value, then prompt " flexibility of user is general, and four limbs are not enough unfolded, and has integrally reached mean level ".
Second of embodiment: calculate the bending angle and corresponding benchmark bending angle between the limbs that are connected first is poor
Value, and prompt the user with posture correction information corresponding with the first difference；And/or each limbs are calculated relative to preset direction
Second difference of orientation angles and corresponding reference bearing angle, and prompt the user with posture correction letter corresponding with the second difference
Breath.
Posture correction information in present embodiment is the correction to user's limbs dimension.For example, being connected between limbs
The first difference that bending angle subtracts corresponding benchmark bending angle is greater than 0, and user is prompted to tighten corresponding limbs；First difference
Less than 0, user is prompted to unfold corresponding limbs.In another example orientation angles and corresponding benchmark side of the limbs relative to preset direction
Second difference of parallactic angle degree is greater than 0, prompts user to the direction activity limbs for reducing the second difference；Second difference is mentioned less than 0
Show user to the direction activity limbs for increasing the second difference.
In the present embodiment, identified from pose presentation by obtaining the pose presentation of user, and using position identification model
The thermodynamic chart at position out, and then the posture information of user is calculated, to obtain the posture information of user by position identification model；
By the posture information and reference attitude information according to user, posture correction information is prompted the user with, so as to compare benchmark
Posture automatically corrects the nonstandard posture of user, that is, realizes the automatic straightening of user's posture, voluntarily compare without user
Correction is corrected without main broadcaster's real-time online, is saved human cost and operation cost, is avoided correcting a mistake or correct not in time
The problem of.
Further, by calculating the orientation of bending angle and/or limbs between the limbs that are connected relative to preset direction
Angle obtains the posture of user, consistent to the requirement of posture with actual life, so that posture correction is more rationally, effectively.So
Afterwards, in conjunction with benchmark bending angle range and/or reference bearing angle range, the score of user is calculated, from bending angle and orientation
The dimension of angle gives a mark to posture, and the accuracy of accurate evaluation user's posture achievees the effect that effectively correct posture.
Example IV
The embodiment of the present invention provides a kind of application scenarios, and detailed description user carries out position identification by terminal and posture is rectified
Positive process.
In the application scenarios of the present embodiment, user opens the posture correcting function of terminal, and selects a kind of posture, such as
Pigeon formula.Terminal opens the area image in camera and acquisition camera pickup area.Camera pickup area, which can be, to be taken the photograph
As the panel region in front of head, should there is no shelter, barrier in the region, be suitble to user movement.Using position identification model
User position in the predeterminable area of area image is identified.Predeterminable area can be the central area of area image, then
Preferentially waist, chest are identified.When recognizing user in predeterminable area, such as the waist of user, chest are recognized,
Then determine that user is located at suitable position, can come into play., whereas if unidentified in predeterminable area arrive user, then root
According to the recognition result in whole region image, moving direction is prompted the user with, so that user is located in predeterminable area.
It is corresponding by display screen playback references posture information when recognizing user in the predeterminable area in area image
Instructional video.Wherein, reference attitude information refers to the corresponding reference attitude information of posture of user's selection.Reference attitude information preservation
In the configuration file of terminal.
After instructional video finishes playing or in playing process, terminal prompts the user with " starting intelligent correction ", then takes the photograph
It is projected on the display screen of terminal and and instructional video as head acquires the pose presentation of user in real time, and by the pose presentation of user
Subregion is shown.
Then, position identification device obtains the pose presentation for showing at least one position of user, and pose presentation is defeated
Enter into position identification model；Obtain the location information at least one position of the user of the position identification model output；Root
According to the location information at least one position, the posture information of user is calculated；According to the posture information of user and reference attitude information,
Prompt the user with posture correction information, such as " arm again high point ", " maintaining very well " etc..
After completing the correction of this posture, terminal show that all previous posture corrects information and corresponding comprehensive score, for
Family comparison is checked, knows rectification effect.
Embodiment five
The present embodiment advanced optimizes above-described embodiment, specifically defines the training process of position identification model.
It specifically, further include position identification mould before the pose presentation for obtaining at least one position for showing user
The training process of type.The training process of position identification model includes: to obtain position identification model to be trained, and it is useful to obtain display
The sample image at least one position of family and location tags information corresponding at least one position, optionally, at least one portion
The corresponding location tags information in position refers to the corresponding position coordinates in position or regional scope.Wherein, regional scope can use coordinate
Range, the label of geometry or thermodynamic chart indicate.
Then, according to sample image location tags information corresponding at least one position, training position identification model.Tool
Body, sample image is input to position identification model to be trained, the parameter in the identification model of iteration position, so that model is defeated
The corresponding location tags information at least one position in the sample image of input is approached out.
Further, according to sample image location tags information corresponding at least one position, training position identifies mould
Type, comprising: according to the corresponding location tags information at least one position, generate the thermodynamic chart at corresponding each position；According to sample graph
The thermodynamic chart of picture and corresponding each position, training position identification model.
Illustratively, 10,000 sample images are acquired by Baidu's crowdsourcing form, each portion is then marked on sample image
Position.By Gaussian Blur algorithm, handled to obtain the corresponding thermodynamic chart in each position, such as head to having markd sample image
The corresponding thermodynamic chart in portion, the corresponding thermodynamic chart of left shoulder, corresponding thermodynamic chart of right shoulder etc..By unlabelled sample image be input to
Trained position identification model, the parameter in the identification model of iteration position, so that model output approaches in the sample image of input
The corresponding thermodynamic chart at least one position.
Embodiment six
Fig. 4 is a kind of structural schematic diagram for position identification device that the embodiment of the present invention six provides, including first obtains mould
Block 41, input module 42, second obtain module 43.
First obtains module 41, for obtaining the pose presentation for showing at least one position of user；
Input module 42, for the pose presentation to be input in the identification model of position；
Second obtains module 43, the position at least one position of the user for obtaining the position identification model output
Information；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction
Submodel includes: that the sequentially connected convolution module of multiple groups and down-sampled module for inputting pose presentation export the posture figure
Feature as in；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, described in inputting
Feature in pose presentation exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling mould
Block.
The embodiment of the invention provides a kind of position identification models, realize a kind of position recognizer based on this model.
Wherein, position identification model includes the feature extraction submodel for feature extraction and the location detection submodule for position identification
Type, feature extraction submodel include the sequentially connected convolution module of multiple groups and down-sampled module, by by sampling module to extraction
To feature zoom in and out, to reduce intrinsic dimensionality, reduce data redundancy；Location detection submodel includes: multiple successively to connect
The detection module and convolution module connect is gradually corrected recognition result, by multiple connected detection modules to reach
Preferable recognition effect；Detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling
Module, rise sampling is converted to high-definition picture to the low-resolution image received, differentiates enough to get to have
The location information of rate, it is seen then that the structure of position identification model is simple, parameter is few, can be stored in terminal and recognition speed is very fast,
And then the position recognizer based on position identification model can satisfy the requirement of real-time of live streaming in terminal operating.
Optionally, position identification model further includes skip floor；Skip floor is by the down-sampled mould of at least one in feature extraction submodel
After the feature of block output is spliced, the feature for rising sampling module output corresponding with location detection submodel is spliced.
Optionally, convolution module includes M bottleneck, and each bottleneck includes the expansion layer being connected, volume
Lamination and compression layer, output channel number are N, and wherein M, N are natural number, 3≤M≤5,16≤N≤96.Each bottleneck packet
Include the point-by-point convolution of depth convolution sum.
Optionally, second module 43 is obtained at least one position for the user for obtaining the position identification model output
When location information, it is specifically used for: obtains the thermodynamic chart of user's corresponding position of the position identification model output.
Optionally, which further includes determining module, is used for: corresponding in the user for obtaining the position identification model output
After the thermodynamic chart at position, from the thermodynamic chart at each position of correspondence, the position coordinates at each position are determined respectively.
Optionally, which further includes computing module and cue module, and computing module is used to obtain the position identification
After the location information at least one position of the user of model output, according to the location information at least one position, meter
Calculate the posture information of the user.Cue module is used for posture information and reference attitude information according to the user, to user
Posture is prompted to correct information.
Optionally, computing module calculates the posture letter of the user in the location information according at least one position
When breath, it is specifically used for: according to the location information at least one position, calculates the bending angle between the connected limbs of user；
And/or the location information according at least one position, calculate orientation angles of each limbs of user relative to preset direction.
Optionally, reference attitude information include: benchmark bending angle range between connected limbs and each limbs relative to
The reference bearing angle range of preset direction；Correspondingly, cue module is in the posture information and reference attitude information according to user,
Prompt the user with posture correction information when, be specifically used for: according between connected limbs bending angle and corresponding benchmark be bent
Angular range calculates the first score of the limbs that are connected；Orientation angles and corresponding base according to each limbs relative to preset direction
Quasi- orientation angles range calculates the second score of each limbs；Summation is weighted to the first score and the second score, obtains user
Comprehensive score；Prompt the user with posture correction information corresponding with comprehensive score.
Optionally, which further includes acquisition module and playing module.Collection model is used to show user's in acquisition
Area image before the pose presentation at least one position, in acquisition camera pickup area.When in the default of area image
When recognizing user in region, pass through the corresponding instructional video of playing module playback references posture information.
Optionally, which further includes training module, and for obtaining position identification model to be trained, it is useful to obtain display
The sample image at least one position of family and location tags information corresponding at least one position, according to sample image and at least
The corresponding location tags information in one position, training position identification model.
Position provided by any embodiment of the invention, which can be performed, in position identification device provided by the embodiment of the present invention knows
Other method has the corresponding functional module of execution method and beneficial effect.
Embodiment seven
Fig. 5 is a kind of structural schematic diagram for terminal that the embodiment of the present invention seven provides, as shown in figure 5, the terminal includes place
Manage device 50, memory 51, input unit 52, output device 53；The quantity of processor 50 can be one or more in terminal, figure
In 5 by taking a processor 50 as an example；Processor 50, memory 51, input unit 52, output device 53 in terminal can pass through
Bus or other modes connect, in Fig. 5 for being connected by bus.
Memory 51 is used as a kind of computer readable storage medium, can be used for storing software program, journey can be performed in computer
Sequence and module, if the corresponding program instruction/module of position recognition methods in the embodiment of the present invention is (for example, position identification dress
First in setting obtains module 41, input module 42, second obtains module 43).Processor 50 is stored in memory by operation
Software program, instruction and module in 51 are realized above-mentioned thereby executing the various function application and data processing of terminal
Position recognition methods.
Memory 51 can mainly include storing program area and storage data area, wherein storing program area can store operation system
Application program needed for system, at least one function；Storage data area, which can be stored, uses created data etc. according to terminal.This
Outside, memory 51 may include high-speed random access memory, can also include nonvolatile memory, for example, at least a magnetic
Disk storage device, flush memory device or other non-volatile solid state memory parts.In some instances, memory 51 can be further
Including the memory remotely located relative to processor 50, these remote memories can pass through network connection to terminal.It is above-mentioned
The example of network includes but is not limited to internet, intranet, local area network, mobile radio communication and combinations thereof.
Input unit 52 can be used for receiving the number or character information of input, and generate with the user setting of terminal and
The related key signals input of function control, such as user select a kind of posture by input unit 52.Output device 53 may include
Display screen etc. shows the audio frequency apparatuses such as equipment and loudspeaker.
Optionally, terminal can also include camera, for obtaining the posture figure for showing at least one position of user
Picture.The camera can be the front camera in terminal.
Embodiment eight
The embodiment of the present invention eight also provides a kind of computer readable storage medium for being stored thereon with computer program, calculates
Machine program is used to execute a kind of position recognition methods when being executed by computer processor, this method comprises:
Obtain the pose presentation for showing at least one position of user；
The pose presentation is input in the identification model of position；
Obtain the location information at least one position of the user of the position identification model output；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction
Submodel includes: that the sequentially connected convolution module of multiple groups and down-sampled module for inputting pose presentation export the posture figure
Feature as in；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, described in inputting
Feature in pose presentation exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling mould
Block.
Certainly, a kind of computer-readable storage medium being stored thereon with computer program provided by the embodiment of the present invention
Matter, computer program are not limited to method operation as above, position provided by any embodiment of the invention can also be performed and know
Relevant operation in other method.
By the description above with respect to embodiment, it is apparent to those skilled in the art that, the present invention
It can be realized by software and required common hardware, naturally it is also possible to which by hardware realization, but in many cases, the former is more
Good embodiment.Based on this understanding, technical solution of the present invention substantially in other words contributes to the prior art
Part can be embodied in the form of software products, which can store in computer readable storage medium
In, floppy disk, read-only memory (Read-Only Memory, ROM), random access memory (Random such as computer
Access Memory, RAM), flash memory (FLASH), hard disk or CD etc., including some instructions are with so that a computer is set
The method that standby (can be personal computer, server or the network equipment etc.) executes each embodiment of the present invention.
It is worth noting that, included each unit and module are only pressed in the embodiment of above-mentioned position identification device
It is divided, but is not limited to the above division according to function logic, as long as corresponding functions can be realized；In addition,
The specific name of each functional unit is also only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The present invention is not limited to specific embodiments here, be able to carry out for a person skilled in the art it is various it is apparent variation, again
Adjustment and substitution are without departing from protection scope of the present invention.Therefore, although by above embodiments to the present invention carried out compared with
For detailed description, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, can be with
Including more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (10)
1. a kind of position recognition methods characterized by comprising
Obtain the pose presentation for showing at least one position of user；
The pose presentation is input in the identification model of position；
Obtain the location information at least one position of the user of the position identification model output；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction submodule
Type includes: that the sequentially connected convolution module of multiple groups and down-sampled module export in the pose presentation for inputting pose presentation
Feature；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, for inputting the posture
Feature in image exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling module.
2. the method according to claim 1, wherein the position identification model further includes skip floor；
After the skip floor is spliced the feature of the down-sampled module output of at least one in feature extraction submodel, examined with position
The corresponding feature for rising sampling module output in submodel is surveyed to be spliced.
3. the method according to claim 1, wherein the convolution module includes M bottleneck, each
Bottleneck includes the expansion layer, convolutional layer and compression layer being connected, and output channel number is N, and wherein M, N are natural numbers, 3≤
M≤5,16≤N≤96.
4. according to the method described in claim 3, it is characterized in that, each bottleneck includes that depth convolution sum is rolled up point by point
Product.
5. the method according to claim 1, wherein the user for obtaining position identification model output
The location information at least one position, comprising:
Obtain the thermodynamic chart of user's corresponding position of the position identification model output.
6. according to the method described in claim 5, it is characterized in that, corresponding in the user for obtaining the position identification model output
After the thermodynamic chart at position, further includes:
From the thermodynamic chart at each position of correspondence, the position coordinates at each position are determined respectively.
7. method according to claim 1-6, which is characterized in that defeated in the acquisition position identification model
After the location information at least one position of user out, further includes:
According to the location information at least one position, the posture information of the user is calculated；
According to the posture information of the user and reference attitude information, posture correction information is prompted the user with.
8. a kind of position identification device characterized by comprising
First obtains module, for obtaining the pose presentation for showing at least one position of user；
Input module, for the pose presentation to be input in the identification model of position；
Second obtains module, the location information at least one position of the user for obtaining the position identification model output；
Wherein, the position identification model includes feature extraction submodel and location detection submodel, the feature extraction submodule
Type includes: that the sequentially connected convolution module of multiple groups and down-sampled module export in the pose presentation for inputting pose presentation
Feature；
The location detection submodel includes: multiple sequentially connected detection modules and convolution module, for inputting the posture
Feature in image exports the location information of corresponding position；
The detection module includes: the sequentially connected convolution module of down-sampled module, multiple groups being connected and liter sampling module.
9. a kind of terminal characterized by comprising
One or more processors；
Memory, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The now position recognition methods as described in any in claim 1-7.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The position recognition methods as described in any in claim 1-7 is realized when execution.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810820840.0A CN109117753B (en) | 2018-07-24 | 2018-07-24 | Part recognition method, device, terminal and storage medium |
| CN202110312615.8A CN112911393B (en) | 2018-07-24 | 2018-07-24 | Method, device, terminal and storage medium for identifying part |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810820840.0A CN109117753B (en) | 2018-07-24 | 2018-07-24 | Part recognition method, device, terminal and storage medium |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110312615.8A Division CN112911393B (en) | 2018-07-24 | 2018-07-24 | Method, device, terminal and storage medium for identifying part |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109117753A true CN109117753A (en) | 2019-01-01 |
| CN109117753B CN109117753B (en) | 2021-04-20 |
Family
ID=64862491
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110312615.8A Active CN112911393B (en) | 2018-07-24 | 2018-07-24 | Method, device, terminal and storage medium for identifying part |
| CN201810820840.0A Active CN109117753B (en) | 2018-07-24 | 2018-07-24 | Part recognition method, device, terminal and storage medium |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110312615.8A Active CN112911393B (en) | 2018-07-24 | 2018-07-24 | Method, device, terminal and storage medium for identifying part |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN112911393B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110139115A (en) * | 2019-04-30 | 2019-08-16 | 广州虎牙信息科技有限公司 | Virtual image attitude control method, device and electronic equipment based on key point |
| CN110264539A (en) * | 2019-06-18 | 2019-09-20 | 北京字节跳动网络技术有限公司 | Image generating method and device |
| CN110443148A (en) * | 2019-07-10 | 2019-11-12 | 广州市讯码通讯科技有限公司 | A kind of action identification method, system and storage medium |
| CN110942005A (en) * | 2019-11-21 | 2020-03-31 | 网易（杭州）网络有限公司 | Object recognition method and device |
| CN111683263A (en) * | 2020-06-08 | 2020-09-18 | 腾讯科技（深圳）有限公司 | Live broadcast guiding method, device, equipment and computer readable storage medium |
| CN113177432A (en) * | 2021-03-16 | 2021-07-27 | 重庆兆光科技股份有限公司 | Head pose estimation method, system, device and medium based on multi-scale lightweight network |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI789974B (en) * | 2021-11-04 | 2023-01-11 | 財團法人資訊工業策進會 | Assistance system and method for guiding exercise postures in live broadcast |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160110592A1 (en) * | 2012-01-13 | 2016-04-21 | C/O Sony Corporation | Image processing device, method and program for moving gesture recognition using difference images |
| US20160171340A1 (en) * | 2014-12-11 | 2016-06-16 | Intel Corporation | Labeling component parts of objects and detecting component properties in imaging data |
| CN105989331A (en) * | 2015-02-11 | 2016-10-05 | 佳能株式会社 | Facial feature extraction apparatus, facial feature extraction method, image processing equipment and image processing method |
| CN107392201A (en) * | 2017-06-09 | 2017-11-24 | 中国科学院自动化研究所 | The pillar recognition methods of catenary mast, storage medium, processing equipment |
| CN107403168A (en) * | 2017-08-07 | 2017-11-28 | 青岛有锁智能科技有限公司 | A kind of facial-recognition security systems |
| CN108062536A (en) * | 2017-12-29 | 2018-05-22 | 纳恩博（北京）科技有限公司 | A kind of detection method and device, computer storage media |
Family Cites Families (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102385695A (en) * | 2010-09-01 | 2012-03-21 | 索尼公司 | Human body three-dimensional posture identifying method and device |
| JP5777367B2 (en) * | 2011-03-29 | 2015-09-09 | キヤノン株式会社 | Pattern identification device, pattern identification method and program |
| KR101907077B1 (en) * | 2012-03-06 | 2018-10-11 | 삼성전자주식회사 | Method and apparatus for motion recognition |
| US9400925B2 (en) * | 2013-11-15 | 2016-07-26 | Facebook, Inc. | Pose-aligned networks for deep attribute modeling |
| US20150243031A1 (en) * | 2014-02-21 | 2015-08-27 | Metaio Gmbh | Method and device for determining at least one object feature of an object comprised in an image |
| CN104200491A (en) * | 2014-08-15 | 2014-12-10 | 浙江省新华医院 | Motion posture correcting system for human body |
| WO2018033154A1 (en) * | 2016-08-19 | 2018-02-22 | 北京市商汤科技开发有限公司 | Gesture control method, device, and electronic apparatus |
| CN106372651B (en) * | 2016-08-22 | 2018-03-06 | 平安科技（深圳）有限公司 | The detection method and device of picture quality |
| CN107953329B (en) * | 2016-10-17 | 2021-06-15 | 中国科学院深圳先进技术研究院 | Object recognition and attitude estimation method and device and mechanical arm grabbing system |
| US10902243B2 (en) * | 2016-10-25 | 2021-01-26 | Deep North, Inc. | Vision based target tracking that distinguishes facial feature targets |
| CN108229268A (en) * | 2016-12-31 | 2018-06-29 | 商汤集团有限公司 | Expression Recognition and convolutional neural networks model training method, device and electronic equipment |
| CN107609541B (en) * | 2017-10-17 | 2020-11-10 | 哈尔滨理工大学 | Human body posture estimation method based on deformable convolution neural network |
| CN107704838B (en) * | 2017-10-19 | 2020-09-25 | 北京旷视科技有限公司 | Target object attribute identification method and device |
| CN107798313A (en) * | 2017-11-22 | 2018-03-13 | 杨晓艳 | A kind of human posture recognition method, device, terminal and storage medium |
| CN107895161B (en) * | 2017-12-22 | 2020-12-11 | 北京奇虎科技有限公司 | Real-time attitude identification method and device based on video data and computing equipment |
| CN108038474B (en) * | 2017-12-28 | 2020-04-14 | 深圳励飞科技有限公司 | Face detection method, convolutional neural network parameter training method, device and medium |
| CN108305283B (en) * | 2018-01-22 | 2020-12-08 | 清华大学 | Human behavior recognition method and device based on depth camera and basic gesture |
| CN108182456B (en) * | 2018-01-23 | 2022-03-18 | 哈工大机器人（合肥）国际创新研究院 | Target detection model based on deep learning and training method thereof |
-
2018
- 2018-07-24 CN CN202110312615.8A patent/CN112911393B/en active Active
- 2018-07-24 CN CN201810820840.0A patent/CN109117753B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160110592A1 (en) * | 2012-01-13 | 2016-04-21 | C/O Sony Corporation | Image processing device, method and program for moving gesture recognition using difference images |
| US20160171340A1 (en) * | 2014-12-11 | 2016-06-16 | Intel Corporation | Labeling component parts of objects and detecting component properties in imaging data |
| CN105989331A (en) * | 2015-02-11 | 2016-10-05 | 佳能株式会社 | Facial feature extraction apparatus, facial feature extraction method, image processing equipment and image processing method |
| CN107392201A (en) * | 2017-06-09 | 2017-11-24 | 中国科学院自动化研究所 | The pillar recognition methods of catenary mast, storage medium, processing equipment |
| CN107403168A (en) * | 2017-08-07 | 2017-11-28 | 青岛有锁智能科技有限公司 | A kind of facial-recognition security systems |
| CN108062536A (en) * | 2017-12-29 | 2018-05-22 | 纳恩博（北京）科技有限公司 | A kind of detection method and device, computer storage media |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110139115A (en) * | 2019-04-30 | 2019-08-16 | 广州虎牙信息科技有限公司 | Virtual image attitude control method, device and electronic equipment based on key point |
| CN110264539A (en) * | 2019-06-18 | 2019-09-20 | 北京字节跳动网络技术有限公司 | Image generating method and device |
| WO2020253716A1 (en) * | 2019-06-18 | 2020-12-24 | 北京字节跳动网络技术有限公司 | Image generation method and device |
| CN110443148A (en) * | 2019-07-10 | 2019-11-12 | 广州市讯码通讯科技有限公司 | A kind of action identification method, system and storage medium |
| CN110443148B (en) * | 2019-07-10 | 2021-10-22 | 广州市讯码通讯科技有限公司 | Action recognition method, system and storage medium |
| CN110942005A (en) * | 2019-11-21 | 2020-03-31 | 网易（杭州）网络有限公司 | Object recognition method and device |
| CN111683263A (en) * | 2020-06-08 | 2020-09-18 | 腾讯科技（深圳）有限公司 | Live broadcast guiding method, device, equipment and computer readable storage medium |
| CN113177432A (en) * | 2021-03-16 | 2021-07-27 | 重庆兆光科技股份有限公司 | Head pose estimation method, system, device and medium based on multi-scale lightweight network |
| CN113177432B (en) * | 2021-03-16 | 2023-08-29 | 重庆兆光科技股份有限公司 | Head posture estimation method, system, equipment and medium based on multi-scale lightweight network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112911393B (en) | 2023-08-01 |
| CN112911393A (en) | 2021-06-04 |
| CN109117753B (en) | 2021-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117753A (en) | Position recognition methods, device, terminal and storage medium | |
| CN109274883A (en) | Posture antidote, device, terminal and storage medium | |
| Rogez et al. | Mocap-guided data augmentation for 3d pose estimation in the wild | |
| CN107609383B (en) | 3D face identity authentication method and device | |
| CN105740780B (en) | Method and device for detecting living human face | |
| CN106355147A (en) | Acquiring method and detecting method of live face head pose detection regression apparatus | |
| CN110544301A (en) | Three-dimensional human body action reconstruction system, method and action training system | |
| JP4951498B2 (en) | Face image recognition device, face image recognition method, face image recognition program, and recording medium recording the program | |
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| JPWO2012077286A1 (en) | Object detection apparatus and object detection method | |
| CN104794449B (en) | Gait energy diagram based on human body HOG features obtains and personal identification method | |
| CN110544302A (en) | Human body action reconstruction system and method based on multi-view vision and action training system | |
| CN105069745A (en) | face-changing system based on common image sensor and enhanced augmented reality technology and method | |
| CN111898571A (en) | Action recognition system and method | |
| CN111582036A (en) | Cross-view-angle person identification method based on shape and posture under wearable device | |
| CN109919128A (en) | Acquisition methods, device and the electronic equipment of control instruction | |
| CN110188179A (en) | Speech-oriented identifies exchange method, device, equipment and medium | |
| CN110334609A (en) | A kind of real-time body-sensing method for catching of intelligence | |
| CN112017212B (en) | Training and tracking method and system of face key point tracking model | |
| CN106156739A (en) | A kind of certificate photo ear detection analyzed based on face mask and extracting method | |
| CN109784215A (en) | A kind of in-vivo detection method and system based on improved optical flow method | |
| CN117238031A (en) | Motion capturing method and system for virtual person | |
| CN111428609A (en) | Human body posture recognition method and system based on deep learning | |
| CN110516638A (en) | A kind of sign Language Recognition Method based on track and random forest | |
| CN109382827A (en) | A kind of robot system and its intelligent memory recognition methods |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |