CN109192314A - A kind of cervical vertebra health score assigning device and its application based on multi-instance learning - Google Patents
A kind of cervical vertebra health score assigning device and its application based on multi-instance learning Download PDFInfo
- Publication number
- CN109192314A CN109192314A CN201810981991.4A CN201810981991A CN109192314A CN 109192314 A CN109192314 A CN 109192314A CN 201810981991 A CN201810981991 A CN 201810981991A CN 109192314 A CN109192314 A CN 109192314A
- Authority
- CN
- China
- Prior art keywords
- cervical vertebra
- score value
- packet
- frame
- indicate
- 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
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Abstract
The cervical vertebra health score assigning device and its application that the present invention relates to a kind of based on multi-instance learning.Cervical vertebra health score assigning device of the present invention based on multi-instance learning extracts cervical vertebra health characteristics from computer vision angle from video information；Learn cervical vertebra health score assigning automatically using easy depth camera machine equipment and machine learning model, not only saves equipment cost and mark cost, but also can accurately be scored in the case where not interfering user the cervical vertebra health of user.
Description
Technical field
The cervical vertebra health score assigning device and its application that the present invention relates to a kind of based on multi-instance learning, belong to cervical vertebra health and comment
The technical field divided.
Background technique
Cervical vertebra health is the problem of people pay close attention to always, sitting posture health evaluating, head pose estimation and more examples
The fields such as learning model can all be related to the assessment of cervical vertebra health.
In the method for sitting posture health evaluating, document " Villanueva M B G, Jonai H, Sotoyama M, et
al.Sitting posture and neck and shoulder muscle activities at different
screen height settings of the visual display terminal[J].Industrial health,
1997,35 (3): 330-336. " describes the active evaluation sitting posture health by recording muscle；Document " Kamiya K, Kudo
M,Nonaka H,et al.Sitting posture analysis by pressure sensors[C]//Pattern
Recognition, 2008.ICPR 2008.19th International Conference on.IEEE, 2008:1-4. " note
It has carried and has judged sitting posture state by pressure sensor；Document " Xu W, Huang M C, Amini N, et al.ecushion:A
textile pressure sensor array design and calibration for sitting posture
Analysis [J] .IEEE Sensors Journal, 2013,13 (10): 3926-3934. " is disclosed to be passed using weaving pressure
Sensor obtains data, and assessment sitting posture health on this basis.The above method can directly or indirectly comment sitting posture health
Estimate, and by special installation, assist detecting by physical means, corresponding warning is proposed in user's sitting posture unhealthy status, is reached
To the purpose of cervical spondylosis prevention and recovering aid.But these special installations often bring inconvenience to the normal office work of user,
It higher cost and is not easy to apply.In addition, these methods can only propose to prompt to ill sitting posture, and it cannot reflect Degree of Ill Condition,
There is its limitation in terms of rehabilitation guide.
In head pose estimation field, head pose estimation is the key that infer position and orientation of the head with respect to camera,
It is played a crucial role in many high-level human face analysis tasks, for example, document " Li C, Zhong F, Qin
X.Accurate 3D head pose estimation with noisy RGBD images[C]//Proceedings of
The 33rd Computer Graphics International.ACM, 2016:37-40 " describe corresponding content.In neck
In vertebra health evaluating, human body attitude feature is captured, the whole physical condition of available user, and head pose to
The cervical vertebra health at family has differentiation effect.In general, only rely on the head pose estimation method of color image to illumination variation,
Partial occlusion and characteristic point missing are very sensitive, and the method for being based only upon depth lacks texture and colouring information, are easy by depth
The influence of noise.
Algorithm of Head Pose Estimation in the prior art reach its maturity can the 6DOF parameter to head estimated in real time
Meter.The 6DOF attitude parameter on head includes that three location parameters and three rotation parameters, physical significance respectively indicate head
The position and head of opposite camera therefore can use head pose parameter reflection cervical vertebra health condition with respect to the direction of camera；Example
Such as, " Li C, Zhong F, Qin X.Accurate 3D head pose estimation with noisy RGBD
images[C]//Proceedings of the 33rd Computer Graphics International.ACM,2016:
37-40 " discloses specific Algorithm of Head Pose Estimation.
Multi-instance learning is a kind of Weakly supervised learning method, and data are provided in the form wrapped, the label of packet be it is positive or negative, often
It include several examples and individual features in a packet, exemplary label then not marks.If include in a packet all shows
The label of example is negative, then the label of the packet is negative；If at least one label is positive in the example for including in a packet, should
The label of packet is positive.
“Andrews S,Tsochantaridis I,Hofmann T.Support vector machines for
multiple-instance learning[C]//Advances in neural information processing
Systems.2003:577-584 " discloses mi-SVM multi-instance learning model, and mi-SVM model obtains one by calculation optimization
A optimal hyperlane separates positive example and negative example, while meeting positive closure and containing at least one positive example, does not include in negative packet
Positive example.
, usually will be by medicine ancillary equipment at present in cervical vertebra health score assigning technical research, medicine ancillary equipment is often
Heaviness can generate a degree of negative effect when use to user, and due to the limitation of physical method, to cervical vertebra health
The accuracy of scoring is not also high.
Summary of the invention
In view of the deficiencies of the prior art, the present invention provides a kind of cervical vertebra health score assigning device based on multi-instance learning.
The present invention also provides a kind of methods for carrying out cervical vertebra health score assigning using above-mentioned apparatus.
Term explanation:
Kinect2: being that Microsoft started to sell second generation Kinect for Windows in China in October, 2014
Inductor, i.e., usually said V2.
The technical solution of the present invention is as follows:
A kind of cervical vertebra health score assigning device based on multi-instance learning, including the Kinect2 being connect with computer；It is described
Kinect2 includes colour imagery shot and depth camera.
A method of cervical vertebra health score assigning being carried out using above-mentioned apparatus, is comprised the following steps that
1) in pretreatment stage, pass through the video sequence under Kinect2 continuous acquisition difference sitting posture, including colored graphic sequence
And depth map sequence；
1.1) using colored graphic sequence and depth map sequence, the 6 of each frame is calculated in real time by Algorithm of Head Pose Estimation
Freedom degree head pose, including three location parameters and three rotation parameters；
1.2) speed of the corresponding 6DOF head pose of each frame is calculated by adjacent two frame；The speed pass through by
The 6DOF attitude parameter of consecutive frame asks difference to obtain；
1.3) using the speed of 6DOF head pose and 6DOF head pose as 12 dimensional features of frame；12 dimensional features
For the initial data feature of frame；
1.4) using 30 minutes continuous videos sequences and it includes the initial data features of all frames as a packet goes forward side by side
Rower note, mark score value are CnDivide system；
2) cluster and example division；
2.1) it using the initial data feature of all frames in all packets as input, is carried out by Meanshift clustering algorithm
Cluster, obtains N number of class and corresponding class central point；
2.2) each frame in packet is subordinated to a class；When dividing example, class-mark belonging to frames all in packet is arranged in
One array；If there is continuing to exceed f1A frame belongs to same class, then generates an example, illustrate that the state continue for f1A frame
More than；
If an example is more than f2A frame is then split into multiple examples, and the frame number for making each example include does not surpass
Cross f2；Each exemplary feature is the class central point of its subordinate class；The class of each example subordinate is the frame subordinate for including in example
Class；So far packet is ready to complete with example；To keep data balancing, it is specified that the frame number that each example includes is no more than f2；
Meanshift method is by given distance parameter, by data clusters to the maximum region of probability density.
The example marked off corresponds to the sub- state of cervical vertebra, and the meaning of step 2.2) is, to the continuous cervical vertebra state in each packet,
Frame with similar behavior form is divided into the sub- state of cervical vertebra.
It 3) is that multiple two-values input score value model by multivalue input score value model partition；
3.1) C respectively with 1,2,3 ...nMultivalue input model to divide threshold value, is divided into C by -1n- 1 two-value inputs mould
Type；Every kind of division constructs a sub-classifier hk(x), k ∈ [1, Cn-1]；X is input data, i.e. the packet of pretreatment stage preparation
Data；When being division threshold value with 1 point, then it is considered as negative packet for 1 point, 2~9 points are considered as positive closure；When with 2 points to divide threshold value, then 1~2
Divide and be considered as negative packet, 3~9 points are considered as positive closure；
3.2) exemplary prediction score value is acquired by sub-classifierI indicates that i-th of packet, j indicate the in i-th of packet
J example, k indicate the serial number of sub-classifier；By Cn- 1 sub- classifiers combination is the classifier of multivalue input:
Optimization method:
Wherein, SiIndicate the score value of i-th of packet；sijIndicate j-th of exemplary practical score value in i-th of packet；xijIndicate i-th
J-th of exemplary feature in a packet；W and b indicates the hyperplane in SVM jointly, for dividing example；ξijFor slack variable, use
In the assumed condition for relaxing multi-instance learning model；C is weight coefficient；
Then exemplary final prediction score value:
5) it is merged using Gauss model, obtains final example score value；
Gauss model is combined with the integrated of sub-classifier, distributes greater weight to the sub-classifier more balanced, it is right
Unbalanced sub-classifier distributes smaller weight, obtains more reasonable Multi-class Classifier.G (x) is gauss of distribution function, is used for
Calculate the weight of each sub-classifier.
Formula (1) is rewritten as formula (4):
Exemplary final prediction score value are as follows:
After pretreatment stage obtains exemplary final score value, the example score value directly obtained using pretreatment when application is i.e.
Can, it does not need to carry out second training；When marking long when video parameter, user can be labeled according to same needs, and reflection difference is commented
Estimate the meaning of score value；Again after labeled data, pre-training only need to be re-started, the scoring of respective sense can be provided.
6) on-line stage, the video sequence of user is shot by Kinect2, and inputs computer；
7) by video sequence according to step 1.1) -1.3) method acquire 12 dimensional features of each frame, according to this feature
With the nearest class of Euclidean distance, i.e. frame and selected class has most similar feature, then divides according to the method in step 2)
Example simultaneously obtains exemplary characteristics, i.e., the feature of each sub- state of cervical vertebra；The nearest class of the Euclidean distance is in step 2)
The class that Meanshift is clustered；
8) the sub- state of cervical vertebra is matched according to the feature of the sub- state of cervical vertebra with the example score value of pretreatment stage, is obtained current
The cervical vertebra health score value of each sub- state of cervical vertebra；The same step 7) of matching way is found and the sub- state feature Euclidean distance of the cervical vertebra
Nearest exemplary characteristics indicate the score value of the sub- state of the cervical vertebra with the exemplary score value；
9) according to cervical vertebra health scoring mechanism, whole cervical vertebra health status score value is calculated；
Formula (6) is cervical vertebra health scoring mechanism,For whole cervical vertebra health score assigning:
Wherein, miIndicate exemplary quantity in i-th of packet,Indicate the score value of i-th of packet；XibIndicate packet BiIn morbid state
Example, i.e. score value are less than the example of α；NiIndicate BiMiddle XibQuantity；liIndicate XibIn BiThe position of middle appearance, std are standard deviation
Function；β and γ is weight coefficient；NiCalculating provided by formula (7)；
Wherein, sum indicates that summing function, sgn indicate jump function.
In cervical vertebra health score assigning, form, which occurs, in the difference of ill cervical vertebra state different influences to cervical vertebra health.Neck
Vertebra health hypothesis should meet: cervical vertebra state is more ill, bigger to negatively affecting caused by whole cervical vertebra health；Ill cervical vertebra state
Duration is longer, and whole cervical vertebra health score value is lower；Ill cervical vertebra state time of occurrence is more concentrated, and the score value of whole cervical vertebra is got over
It is low.
Formula (6) is constituted by three, first item BiIn the sum of all example score values, asBasic score value, indicate
Be each healthy example and ill example to entire effect caused by cervical vertebra health；Section 2 is ill cervical vertebra state decaying
, indicate ill cervical vertebra state additional effect caused by holistic health, the ill cervical vertebra example in a packet is more, the packet
Score value is lower, and NiScore value decaying between relationship be not it is linear, use quadratic function indicate NiBetween score value decaying
Relationship.Section 3 is that ill cervical vertebra state is distributed item, and for a certain number of ill examples, distribution more disperses, and standard deviation is got over
Greatly, the decaying of score value caused by is answered smaller；This scoring mechanism be not limited to standard multi-instance learning it is assumed that conforming better to neck
The actual conditions of vertebra health score assigning.
The invention has the benefit that
1. the cervical vertebra health score assigning device of the present invention based on multi-instance learning, features simple structure are easily achieved, are only instructing
Practicing the stage simply marks the long time series of cervical vertebra exercise data, can estimate the health score assigning of cervical vertebra state in short-term, realize user
The real-time scoring of cervical vertebra health；Effectively overcome the difficult problem with medicine ancillary equipment heaviness of cervical vertebra health data mark；
2. the cervical vertebra health score assigning device of the present invention based on multi-instance learning, it is only necessary to a Kinect2 depth camera
Machine can be realized, and it is cheap, convenient to have the advantages that；Using multi-instance learning model, only carried out on cervical vertebra status switch when long
Score value mark, improves work efficiency；Using the packet scoring mechanism for meeting cervical vertebra health assumed condition, the cervical vertebra being calculated is strong
Health score value is more accurate；
3. the cervical vertebra health score assigning device of the present invention based on multi-instance learning, from computer vision angle, from
Cervical vertebra health characteristics are extracted in video information；Learn cervical vertebra automatically using easy depth camera machine equipment and machine learning model
Health score assigning had not only saved equipment cost and mark cost, but also can be strong to the cervical vertebra of user in the case where not interfering user
Kang Jinhang accurately scores.
Detailed description of the invention
Fig. 1 is the method flow diagram of the present invention for carrying out cervical vertebra health score assigning；
Fig. 2 is the structural schematic diagram of the cervical vertebra health score assigning device of the present invention based on multi-instance learning.
Specific embodiment
Below with reference to embodiment and Figure of description, the present invention will be further described, but not limited to this.
Embodiment 1
As shown in Figure 2.
A kind of cervical vertebra health score assigning device based on multi-instance learning, including the Kinect2 being connect with computer；It is described
Kinect2 includes colour imagery shot and depth camera.Computer is connected with Kinect2 depth camera by data line.
Embodiment 2
As shown in Figure 1.
A method of cervical vertebra health score assigning being carried out using 1 described device of embodiment, is comprised the following steps that
1) in pretreatment stage, pass through the video sequence under Kinect2 continuous acquisition difference sitting posture, including colored graphic sequence
And depth map sequence；
1.1) using colored graphic sequence and depth map sequence, the 6 of each frame is calculated in real time by Algorithm of Head Pose Estimation
Freedom degree head pose, including three location parameters and three rotation parameters；
1.2) speed of the corresponding 6DOF head pose of each frame is calculated by adjacent two frame；The speed pass through by
The 6DOF attitude parameter of consecutive frame asks difference to obtain；
1.3) using the speed of 6DOF head pose and 6DOF head pose as 12 dimensional features of frame；12 dimensional features
For the initial data feature of frame；
1.4) using 30 minutes continuous videos sequences and it includes the initial data features of all frames as a packet goes forward side by side
Rower note, mark score value are CnDivide system；Cn=9；Holistic health degree when mark according to user's cervical vertebra in colored graphic sequence,
Middle cervical vertebra health degree is by specialized medical human assessment, and the integral status of cromogram signaling user cervical vertebra is more healthy, and score value is higher,
Score value is marked at 1~9 point；
2) cluster and example division；
2.1) it using the initial data feature of all frames in all packets as input, is carried out by Meanshift clustering algorithm
Cluster, obtains N number of class and corresponding class central point；
2.2) each frame in packet is subordinated to a class；When dividing example, class-mark belonging to frames all in packet is arranged in
One array；If there is continuing to exceed f1A frame belongs to same class, then generates an example, illustrate that the state continue for f1A frame
More than；
If an example is more than f2A frame is then split into multiple examples, and the frame number for making each example include does not surpass
Cross f2；Each exemplary feature is the class central point of its subordinate class；The class of each example subordinate is the frame subordinate for including in example
Class；So far packet is ready to complete with example；To keep data balancing, it is specified that the frame number that each example includes is no more than f2；
Meanshift method is by given distance parameter, by data clusters to the maximum region of probability density.
The example marked off corresponds to the sub- state of cervical vertebra, and the meaning of step 2.2) is, to the continuous cervical vertebra state in each packet,
Frame with similar behavior form is divided into the sub- state of cervical vertebra.
It 3) is that multiple two-values input score value model by multivalue input score value model partition；
3.1) C respectively with 1,2,3 ...nMultivalue input model to divide threshold value, is divided into C by -1n- 1 two-value inputs mould
Type；Every kind of division constructs a sub-classifier hk(x), k ∈ [1, Cn-1]；X is input data, i.e. the packet of pretreatment stage preparation
Data；When being division threshold value with 1 point, then it is considered as negative packet for 1 point, 2~9 points are considered as positive closure；It to divide threshold value, is then regarded for 2 points when with 2 points
Be negative packet, and 1,3~9 points are considered as positive closure；
3.2) exemplary prediction score value is acquired by sub-classifierI indicates that i-th of packet, j indicate the in i-th of packet
J example, k indicate the serial number of sub-classifier；By Cn- 1 sub- classifiers combination is the classifier of multivalue input:
Optimization method:
Wherein, SiIndicate the score value of i-th of packet；sijIndicate j-th of exemplary practical score value in i-th of packet；xijIndicate i-th
J-th of exemplary feature in a packet；W and b indicates the hyperplane in SVM jointly, for dividing example；ξijFor slack variable, use
In the assumed condition for relaxing multi-instance learning model；C is weight coefficient；In the present embodiment, exemplary score value passes through " Andrews
S,Tsochantaridis I,Hofmann T.Support vector machines for multiple-instance
learning[C]//Advances in neural information processing systems.2003:577-584”
Disclosed in multi-instance learning model mi-SVM realize.
Then exemplary final prediction score value:
5) it is merged using Gauss model, obtains final example score value；
Gauss model is combined with the integrated of sub-classifier, distributes greater weight to the sub-classifier more balanced, it is right
Unbalanced sub-classifier distributes smaller weight, obtains more reasonable Multi-class Classifier.G (x) is gauss of distribution function, is used for
Calculate the weight of each sub-classifier.
Formula (1) is rewritten as formula (4):
Exemplary final prediction score value are as follows:
After pretreatment stage obtains exemplary final score value, the example score value directly obtained using pretreatment when application is i.e.
Can, it does not need to carry out second training；When marking long when video parameter, user can be labeled according to same needs, and reflection difference is commented
Estimate the meaning of score value；Again after labeled data, pre-training only need to be re-started, the scoring of respective sense can be provided.
6) on-line stage, the video sequence of user is shot by Kinect2, and inputs computer；
7) by video sequence according to step 1.1) -1.3) method acquire 12 dimensional features of each frame, according to this feature
With the nearest class of Euclidean distance, i.e. frame and selected class has most similar feature, then divides according to the method in step 2)
Example simultaneously obtains exemplary characteristics, i.e., the feature of each sub- state of cervical vertebra；The nearest class of the Euclidean distance is in step 2)
The class that Meanshift is clustered；
8) the sub- state of cervical vertebra is matched according to the feature of the sub- state of cervical vertebra with the example score value of pretreatment stage, is obtained current
The cervical vertebra health score value of each sub- state of cervical vertebra；The same step 7) of matching way is found and the sub- state feature Euclidean distance of the cervical vertebra
Nearest exemplary characteristics indicate the score value of the sub- state of the cervical vertebra with the exemplary score value；
9) according to cervical vertebra health scoring mechanism, whole cervical vertebra health status score value is calculated；
Formula (6) is cervical vertebra health scoring mechanism,For whole cervical vertebra health score assigning:
Wherein, miIndicate exemplary quantity in i-th of packet,Indicate the score value of i-th of packet；XibIndicate packet BiIn morbid state
Example, i.e. score value are less than the example of α；NiIndicate BiMiddle XibQuantity；liIndicate XibIn BiThe position of middle appearance, std are standard deviation
Function；β and γ is weight coefficient；NiCalculating provided by formula (7)；
Wherein, sum indicates that summing function, sgn indicate jump function.
In cervical vertebra health score assigning, form, which occurs, in the difference of ill cervical vertebra state different influences to cervical vertebra health.Neck
Vertebra health hypothesis should meet: cervical vertebra state is more ill, bigger to negatively affecting caused by whole cervical vertebra health；Ill cervical vertebra state
Duration is longer, and whole cervical vertebra health score value is lower；Ill cervical vertebra state time of occurrence is more concentrated, and the score value of whole cervical vertebra is got over
It is low.
Formula (6) is constituted by three, first item BiIn the sum of all example score values, asBasic score value, indicate
Be each healthy example and ill example to entire effect caused by cervical vertebra health；Section 2 is ill cervical vertebra state decaying
, indicate ill cervical vertebra state additional effect caused by holistic health, the ill cervical vertebra example in a packet is more, the packet
Score value is lower, and NiScore value decaying between relationship be not it is linear, use quadratic function indicate NiBetween score value decaying
Relationship.Section 3 is that ill cervical vertebra state is distributed item, and for a certain number of ill examples, distribution more disperses, and standard deviation is got over
Greatly, the decaying of score value caused by is answered smaller；This scoring mechanism be not limited to standard multi-instance learning it is assumed that conforming better to neck
The actual conditions of vertebra health score assigning.
Claims (2)
1. a kind of cervical vertebra health score assigning device based on multi-instance learning, which is characterized in that including what is connect with computer
Kinect2；The Kinect2 includes colour imagery shot and depth camera.
2. a kind of method for carrying out cervical vertebra health score assigning using claim 1 described device, which is characterized in that such as including step
Under:
1) in pretreatment stage, pass through the video sequence under Kinect2 continuous acquisition difference sitting posture, including colored graphic sequence and depth
Spend graphic sequence；
1.1) using colored graphic sequence and depth map sequence, calculate each frame in real time by Algorithm of Head Pose Estimation 6 are free
Spend head pose, including three location parameters and three rotation parameters；
1.2) speed of the corresponding 6DOF head pose of each frame is calculated by adjacent two frame；
1.3) using the speed of 6DOF head pose and 6DOF head pose as 12 dimensional features of frame；12 dimensional features are frame
Initial data feature；
1.4) using 30 minutes continuous videos sequences and it includes the initial data features of all frames as a packet goes forward side by side rower
Note, mark score value are CnDivide system；
2) cluster and example division；
2.1) it using the initial data feature of all frames in all packets as input, is clustered by Meanshift clustering algorithm,
Obtain N number of class and corresponding class central point；
2.2) each frame in packet is subordinated to a class；When dividing example, class-mark belonging to frames all in packet is arranged in one
Array；If there is continuing to exceed f1A frame belongs to same class, then generates an example, illustrate that the state continue for f1It is more than a frame；
If an example is more than f2A frame is then split into multiple examples, and the frame number for making each example include is no more than f2；
Each exemplary feature is the class central point of its subordinate class；The class of each example subordinate is the class for the frame subordinate for including in example；
So far packet is ready to complete with example；
It 3) is that multiple two-values input score value model by multivalue input score value model partition；
3.1) C respectively with 1,2,3 ...nMultivalue input model to divide threshold value, is divided into C by -1n- 1 two-value input model；Often
Kind divides one sub-classifier h of constructionk(x), k ∈ [1, Cn-1]；X is input data, i.e. the bag data of pretreatment stage preparation；
3.2) exemplary prediction score value is acquired by sub-classifierI indicates that i-th of packet, j indicate j-th in i-th of packet to show
Example, k indicate the serial number of sub-classifier；By Cn- 1 sub- classifiers combination is the classifier of multivalue input:
Optimization method:
Wherein, SiIndicate the score value of i-th of packet；sijIndicate j-th of exemplary practical score value in i-th of packet；xijIndicate i-th of packet
In j-th of exemplary feature；W and b indicates the hyperplane in SVM jointly, for dividing example；ξijFor slack variable, for putting
The assumed condition of wide multi-instance learning model；C is weight coefficient；
Then exemplary final prediction score value:
5) it is merged using Gauss model, obtains final example score value；
Formula (1) is rewritten as formula (4):
Exemplary final prediction score value are as follows:
6) on-line stage, the video sequence of user is shot by Kinect2, and inputs computer；
7) by video sequence according to step 1.1) -1.3) method acquire 12 dimensional features of each frame, Europe is matched according to this feature
Formula is apart from nearest class, i.e., the frame and selected class have most similar feature, then divides example according to the method in step 2)
And obtain exemplary characteristics, i.e., the feature of each sub- state of cervical vertebra；
8) the sub- state of cervical vertebra is matched according to the feature of the sub- state of cervical vertebra with the example score value of pretreatment stage, is obtained current each
The cervical vertebra health score value of the sub- state of cervical vertebra；The same step 7) of matching way is found nearest with the sub- state feature Euclidean distance of the cervical vertebra
Exemplary characteristics, the score value of the sub- state of the cervical vertebra is indicated with the exemplary score value；
9) according to cervical vertebra health scoring mechanism, whole cervical vertebra health status score value is calculated；
Formula (6) is cervical vertebra health scoring mechanism,For whole cervical vertebra health score assigning:
Wherein, miIndicate exemplary quantity in i-th of packet,Indicate the score value of i-th of packet；XibIndicate packet BiIn ill example,
That is example of the score value less than α；NiIndicate BiMiddle XibQuantity；liIndicate XibIn BiThe position of middle appearance, std are standard deviation function；
β and γ is weight coefficient；NiCalculating provided by formula (7)；
Wherein, sum indicates that summing function, sgn indicate jump function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810981991.4A CN109192314B (en) | 2018-08-27 | 2018-08-27 | Cervical vertebra health scoring device based on multi-example learning and application thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810981991.4A CN109192314B (en) | 2018-08-27 | 2018-08-27 | Cervical vertebra health scoring device based on multi-example learning and application thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109192314A true CN109192314A (en) | 2019-01-11 |
| CN109192314B CN109192314B (en) | 2021-08-31 |
Family
ID=64916320
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810981991.4A Active CN109192314B (en) | 2018-08-27 | 2018-08-27 | Cervical vertebra health scoring device based on multi-example learning and application thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109192314B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110507336A (en) * | 2019-07-23 | 2019-11-29 | 广东省医疗器械研究所 | A kind of personalized method for cervical vertebra monitoring and correction |
| CN113782217A (en) * | 2021-09-16 | 2021-12-10 | 工银科技有限公司 | Human health condition grading method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250924A (en) * | 2016-07-27 | 2016-12-21 | 南京大学 | A kind of newly-increased category detection method based on multi-instance learning |
| US20180046941A1 (en) * | 2016-08-12 | 2018-02-15 | Qualcomm Incorporated | Systems and methods for multi-instance learning-based classification for streaming inputs |
| CN207441258U (en) * | 2017-07-03 | 2018-06-01 | 广东工业大学 | A kind of rehabilitation training assessment system based on somatosensory device |
| CN108154932A (en) * | 2017-12-28 | 2018-06-12 | 大连医诚医用科技成果转移转化有限公司 | examination method based on Kinect |
-
2018
- 2018-08-27 CN CN201810981991.4A patent/CN109192314B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250924A (en) * | 2016-07-27 | 2016-12-21 | 南京大学 | A kind of newly-increased category detection method based on multi-instance learning |
| US20180046941A1 (en) * | 2016-08-12 | 2018-02-15 | Qualcomm Incorporated | Systems and methods for multi-instance learning-based classification for streaming inputs |
| CN207441258U (en) * | 2017-07-03 | 2018-06-01 | 广东工业大学 | A kind of rehabilitation training assessment system based on somatosensory device |
| CN108154932A (en) * | 2017-12-28 | 2018-06-12 | 大连医诚医用科技成果转移转化有限公司 | examination method based on Kinect |
Non-Patent Citations (2)
| Title |
|---|
| ANDREWS S等: "Support vector machines for multiple-instance learning", 《PROCEEDINGS OF THE 15TH INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS 》 * |
| 郑传钦: "基于Kinect的颈部康复训练系统的设计与实现", 《电子技术与软件工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110507336A (en) * | 2019-07-23 | 2019-11-29 | 广东省医疗器械研究所 | A kind of personalized method for cervical vertebra monitoring and correction |
| CN113782217A (en) * | 2021-09-16 | 2021-12-10 | 工银科技有限公司 | Human health condition grading method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109192314B (en) | 2021-08-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108460345A (en) | A kind of facial fatigue detection method based on face key point location | |
| CN108664947A (en) | A kind of fatigue driving method for early warning based on Expression Recognition | |
| WO2018120964A1 (en) | Posture correction method based on depth information and skeleton information | |
| CN110084156A (en) | A kind of gait feature abstracting method and pedestrian's personal identification method based on gait feature | |
| CN104123549B (en) | Eye positioning method for real-time monitoring of fatigue driving | |
| CN109948472A (en) | A kind of non-intrusion type human thermal comfort detection method and system based on Attitude estimation | |
| CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
| CN109657609A (en) | Face identification method and system | |
| CN109840565A (en) | A kind of blink detection method based on eye contour feature point aspect ratio | |
| CN110222665A (en) | Human motion recognition method in a kind of monitoring based on deep learning and Attitude estimation | |
| CN105487665A (en) | Method for controlling intelligent mobile service robot based on head posture recognition | |
| CN109002783A (en) | Rescue the human testing in environment and gesture recognition method | |
| CN110188701A (en) | Dress ornament recognition methods, system and terminal based on the prediction of human body key node | |
| CN101777116A (en) | Method for analyzing facial expressions on basis of motion tracking | |
| CN110020632A (en) | A method of the recognition of face based on deep learning is for detecting fatigue driving | |
| CN103902992B (en) | Human face recognition method | |
| CN106529502B (en) | Lip reading recognition methods and device | |
| CN109902565A (en) | The Human bodys' response method of multiple features fusion | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| CN104794446B (en) | Human motion recognition method and system based on synthesis description | |
| CN109344694A (en) | A kind of human body elemental motion real-time identification method based on three-dimensional human skeleton | |
| CN102567716A (en) | Face synthetic system and implementation method | |
| CN109192314A (en) | A kind of cervical vertebra health score assigning device and its application based on multi-instance learning | |
| CN101539989A (en) | Human face detection-based method for testing incorrect reading posture | |
| CN108830223A (en) | A kind of micro- expression recognition method based on batch mode Active Learning |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Qin Xueying Inventor after: Li Jiachen Inventor after: Xu Songhua Inventor before: Li Jiachen Inventor before: Qin Xueying Inventor before: Xu Songhua |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |