A kind of multi-mode Human bodys' response method based on sport biomechanics
Technical field
The invention belongs to Human bodys' response technical fields, and in particular to a kind of multi-mode people based on sport biomechanics
Body Activity recognition method.
Background technique
Sport biomechanics technology has obtained extensively in fields such as ergonomics, training, medical diagnosis and rehabilitation trainings
General concern and application, but the less use in the researchs such as human body daily behavior identification.Existing research is mainly according to sensor number
According to time domain and the discrete features such as frequency domain carry out action recognition, less focus on spy of the human action in space, time, power effect
Property, the expression of movement is not enough, classify it is not careful enough.Therefore, sensor-based continuous multi-mode behavior is known
There are many more problems to need to refine and further investigate.
Summary of the invention
The multi-mode Human bodys' response method based on sport biomechanics that the purpose of the present invention is to provide a kind of, with people
Body skeleton model is carrier, posture sequence is matching object, biomechanical characterization is identification parameter, establishes and is based on movement biomechanics
The method system of multi-mode Activity recognition for sensor-based multi-mode complexity Human bodys' response technique study and is answered
With offer new thinking and theoretical foundation.
To achieve the above object the invention adopts the following technical scheme:
A kind of multi-mode Human bodys' response method based on sport biomechanics, includes the following steps:
Step 1, based on sport biomechanics, organization of human body is converted into the skeleton with kinematic feature factor
Action model, and determine action description parameter, and then Computational biomechanics feature, and describe human action with these characteristic parameters
Posture establishes the relational model between parameter and action mode;
Step 2, the human action Posture description method based on biomethanics in step 1 analyze human action time series data
The characteristics of, simplify extraction method of key frame of the algorithm in conjunction with frame abatement algorithm using based on curve, is continuously moved in multi-mode
Make carry out border detection, realize elemental motion and the segmentation of transitional movement and automatic marking;
The characteristics of step 3, analysis elemental motion mode and transitional movement mode, is established substantially dynamic using multi-model thought
The identification model of operation mode and transitional movement mode realizes the identification of multi-mode behavior.
As a further solution of the present invention, the step 1 the following steps are included:
Step 1.1, the motion information for extracting key node are modeled, the point, line, surface geometric element that artis is formed
Between relative tertiary location and its variation as act content representation;
Step 1.2, by relative translation amount between artis in initial data, artis three-dimensional coordinate, acceleration, angular speed,
The bottom datas such as Eulerian angles are converted to the characteristic parameter with description body biomechanics characteristic, and are retouched with these characteristic parameters
Human action posture is stated, the relational model between parameter and action mode is established.
As a further solution of the present invention, the step 2 the following steps are included:
Step 2.1, action mode can be expressed by the crucial posture in posture sequence continuous in human motion, to adopting
Human posture's sequence of collection carries out key frame primary election, determines operation limit frame, constructs candidate key-frames set;
Step 2.2, using operation limit frame as the segmentation foundation of action data, determine elemental motion posture sequence data with
Transitional movement posture sequence data;
Step 2.3 further cuts down environment in data acquisition, capture device noise pair by key frame abatement algorithm
The influence of data, " abatement " redundancy key frames.
As a further solution of the present invention, the step 3 the following steps are included:
Step 3.1 analyzes different elemental motion modes using the multivariate Gaussian distribution inspection method based on pivot analysis
Data characteristics determine the distribution of data;
Step 3.2, according to the data distribution in step 3.1, select suitable statistical method to extract action mode characteristic, point
The identification model of all kinds of elemental motion modes is not established;
Step 3.3, for the modeling problem of transitional movement mode, consider that transitional movement mode has complicated dynamic special
Property, extracting relation of variables using difference changes, and is combined with multistage submodel and is modeled, with accurate description transitional movement mode
Dynamic characteristic.
As a further solution of the present invention, the step 3.2 specifically: set the sample data of certain class elemental motion as Xst
(N × J), N are number of samples, and J is variable number, sample covariance matrix Rst(J × J) analyzes to obtain corresponding unit through PCA
Orthogonal vectors P (J × J)=[p1,p2,...,pJ], meet Ast=PTRstP, wherein diagonal matrix Ast=diag (λ1,λ2,...
λJ) it is RstCharacteristic value, then j principal component vector Tst(N × J)=[t1,t2,...,tJ], tj=XstPj, Pj(J × 1) is load
J-th of column vector of matrix P can be obtained by the property that multivariate Gaussian is distributed: if the equal Gaussian distributed of J principal component, X
Approximation obeys J member Gaussian Profile.
The beneficial effects of the present invention are: a kind of multi-mode Human bodys' response method based on sport biomechanics is invented,
For human motion and the continuity and periodic feature of motion capture data sequence, based on sport biomechanics, by people
Body structure is converted to the skeleton action model with kinematic feature factor, and describes human action appearance with these characteristic parameters
Gesture establishes the relational model between parameter and action mode;The characteristics of by analysis human action time series data, using based on song
Line simplifies extraction method of key frame of the algorithm in conjunction with frame abatement algorithm, carries out border detection, reality to continuous action in multi-mode
Existing elemental motion and the segmentation of transitional movement and automatic marking.Secondly, establishing elemental motion mode and mistake using multi-model thought
The identification model of action mode is crossed, realizes the identification of multi-mode behavior.The present invention is using human skeleton model as carrier, posture sequence
It is identification parameter for matching object, biomechanical characterization, the method for establishing the multi-mode Activity recognition based on sport biomechanics
System, for sensor-based multi-mode complexity Human bodys' response technique study and application provide new thinking and it is theoretical according to
According to.
Detailed description of the invention
Fig. 1 is the multi-mode behavior in a kind of multi-mode Human bodys' response method based on sport biomechanics of the present invention
Recognition methods system;
Fig. 2 is human action bone in a kind of multi-mode Human bodys' response method based on sport biomechanics of the present invention
Illustraton of model.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Whole elaboration.
As shown in Figure 1, a kind of multi-mode Human bodys' response method based on sport biomechanics, includes the following steps:
Step 1, based on sport biomechanics, organization of human body is converted into the skeleton with kinematic feature factor
Action model, and determine action description parameter, and then Computational biomechanics feature.And human action is described with these characteristic parameters
Posture is established the relational model between parameter and action mode, is specifically implemented according to the following steps:
22 as shown in Figure 2 articulation nodes of step 1.1, selection, extract point, line, surface geometric element that joint is formed it
Between relative tertiary location and its variation as act content representation;
The point, line, surface geometric element set that joint is formed is that the minimum of the corresponding regional area of different action modes is constituted
Unit, the characteristic set that relative positional relationship and its variation between minimum Component units are formed can express extensive movement mould
Formula;Angle between point, line, surface geometric element and apart from etc. reflect space between minimum Component units from different perspectives
Positional relationship can more fully reflect the information of different action modes;The angular speed and acceleration signature in joint are directly anti-
The cadence information for having reflected movement, can efficiently differentiate the different action classification of rhythm.
Step 1.2: by relative translation amount between artis in initial data, artis three-dimensional coordinate, acceleration, angular speed,
The bottom datas such as Eulerian angles are converted to the characteristic parameter with description body biomechanics characteristic, and are retouched with these characteristic parameters
Human action posture is stated, the relational model between parameter and action mode is established;
The geometric element set of posture is made of the point, line, surface that joint is formed, opposite between geometric element in order to describe
Spatial position and its variation select the characteristic parameter of 9 seed types as the space-time characteristic of human action posture:
Joint adjust the distance feature, joint and bone distance feature, joint and plan range feature, bone to angle feature,
Bone and plane included angle feature, plane and plane included angle feature, joint hyperspin feature, joint angle velocity characteristic, joint velocity
Feature.
Step 2, the human action Posture description method based on biomethanics in step 1 analyze human action time series data
The characteristics of, simplify extraction method of key frame of the algorithm in conjunction with frame abatement algorithm using based on curve, is continuously moved in multi-mode
Make carry out border detection, realize elemental motion and the segmentation of transitional movement and automatic marking, is specifically implemented according to the following steps:
Step 2.1, action mode can be expressed by the crucial posture in posture sequence continuous in human motion.Human body
The motion profile in each joint is regarded as the full curve in space, extracts the extreme point on curve and is used as movement " boundary " posture, right
Human posture's sequence application curves of acquisition simplify algorithm, determine operation limit frame, construct candidate key-frames set.
Step 2.2, using operation limit frame F as the segmentation foundation of action data mode, determine elemental motion posture sequence
Data and transitional movement posture sequence data;Movement starting keyframe F in candidate key posture frame sequenceStartIt is terminated with movement
Key frame FEndAs the segmentation foundation of elemental motion mode, then elemental motion is FStartWith FEndBetween action data.Transition
Action data can terminate key frame F by movement previous in two elemental motionsEndWith the starting keyframe F of latter actionStartReally
It is fixed.
Step 2.3 further cuts down environment in data acquisition, capture device noise pair by key frame abatement algorithm
The influence of data, " abatement " redundancy key frames;
The Euclidean distance of the linear interpolation frame and primitive frame of using consecutive frame is deleted every time as the importance measures factor of frame
Frame with minimum importance, the importance for then updating remaining each frame are further continued for cutting down;And by the satisfactory residue of quantity
Frame is as key frame.
The characteristics of step 3, analysis elemental motion mode and transitional movement mode, is established substantially dynamic using multi-model thought
The identification model of operation mode and transitional movement mode is realized the identification of multi-mode behavior, is specifically implemented according to the following steps:
Step 3.1 analyzes different elemental motion modes using the multivariate Gaussian distribution inspection method based on pivot analysis
Data characteristics determine the distribution of data;
If the sample data of certain class elemental motion is Xst(N × J), N are number of samples, and J is variable number.Sample covariance
Battle array is Rst(J×J).It analyzes to obtain corresponding unit orthogonal vectors P (J × J)=[p through PCA1,p2,...,pJ], meet Ast=
PTRstP, wherein diagonal matrix Ast=diag (λ1,λ2,...λJ) it is RstCharacteristic value.Then j principal component vector Tst(N × J)=
[t1,t2,...,tJ], tj=XstPj, Pj(J × 1) is j-th of column vector of load matrix P.It can by the property that multivariate Gaussian is distributed
: if the equal Gaussian distributed of J principal component, X approximation obeys J member Gaussian Profile.
Step 3.2, according to the data distribution in step 3.1, select suitable statistical method to extract action mode characteristic, if
Elemental motion data are approximate multivariate Gaussian distributions, establish monitoring model using PCA method;If elemental motion data are and only
It is non-gaussian distribution, then order of information is further extracted using ICA;If not only containing Gauss information, but also contain non-Gauss information, then adopts
Order of information is extracted simultaneously with ICA-PCA combination algorithm and low order information is analyzed;
Step 3.3, for the modeling problem of transitional movement mode, consider that transitional movement mode has complicated dynamic special
Property, extracting relation of variables using difference changes, and is combined with multistage submodel and is modeled, with accurate description transitional movement mode
Dynamic characteristic.
The above is present pre-ferred embodiments, for the ordinary skill in the art, according to the present invention
Introduction, in the case where not departing from the principle of the present invention and spirit, changes, modifications, replacement and change that embodiment is carried out
Type is still fallen within protection scope of the present invention.