CN1924932A - Method for correcting noises and errors in human sports trapped data - Google Patents

Abstract

This invention discloses one human motion noise and error data catching correction method, which comprises the following steps: establishing human geometry mode; setting human quality distribution parameters; inputting human motion data; making parameters of human motion joint; establishing mixture Newton-Euler rigid dynamic force equation based on Euler angle and four element number; establishing physical binds optimization mode; setting optimization overlap times; using secondary program method to solve optimization mode to get human motion data.

Description

A kind of method that noise in the human sports trapped data and error are revised

Technical field

The present invention relates to catching of human body movement data, particularly a kind of method that noise in the human sports trapped data and error are revised.

Background technology

The human motion capture technology is a kind ofly can directly catch human action, with these actions of numeral, and the technology of utilizing computing machine that exercise data is handled.It all has important use aspect a lot.In the computer animation field, can generate human body animation true to nature by the capture movement data; At medical domain, calculate each joint of human body suffered power and moment by motion capture data, and then analyze the physiological situation in each joint.Equally, whether reasonable, how to improve action if can analyze athletic motion according to speed, acceleration, the power calculated in the sports field.

In the human motion capture technology, the present the widest human motion capture method that is based on optics of usable range.This method is attached to the track of human synovial position monumented point by tracking, and then calculates the rotational angle (cradle head) or the translational movement (translation joint) in joint.But the human motion capture method based on optics has its limitation: when the position that monumented point pastes is equipped with skew with respect to normal bit or because blocking of human body limb causes some monumented point can't be observed then, the exercise data of catching just has bigger error; In addition, optical device is responsive to external environment condition, even the monumented point subsides is fine, also from not blocking phenomenon, the exercise data of catching still contains noise simultaneously.When the human motion capture technology was applied to medical science and sports field, the exercise data that requires to catch satisfied physical constraint, just satisfies the rigid multibody dynamics equation.Because the sum of errors noise of outwardness makes the rigid multibody dynamics equation to satisfy, therefore need a kind of method to eliminate these sum of errors noises.

The rigid multibody dynamics equation is the physical equation that a batch total is calculated speed, acceleration and the suffered generalized force of each rigid body in the multi-rigid-body system.The rigid multibody dynamics equation has a lot of forms of equal value in theory, but mainly is with the newton-Eulerian equation with better numerical evaluation performance in engineering calculation.

In newton-Eulerian equation,, therefore need some special methods to describe because rotatablely moving of joint is not a kind of vector.Be usually used in describing the method that the joint rotatablely moves and have two kinds:

1) Eulerian angle.Eulerian angle are to use one group to describe the method for joint motions around the corner of X, Y, the rotation of Z axle successively.A rotary joint needs only 3 corners at most and just can describe its rotation fully.The minimum joint that needs the individual corner of N (N=1,2,3) could describe its rotation is called the joint with N degree of freedom.Using Eulerian angle to describe the rotary joint biggest advantage is not have redundancy, can describe its rotation with minimum variable according to the degree of freedom in joint.But an important disadvantages is that Eulerian angle may exist singularity, can not guarantee promptly that also Eulerian angle change in time continuously.

2) unit quaternion.Unit quaternion is that a mould length is 1 four-tuple q=(q ₀q ₁q ₂q ₃) $({\mathrm{<figure-callout\; id="0"\; label="q"\; filenames="A20061011307100161.png,A20061011307100162.png"\; state="\{\{state\}\}">q</figure-callout>}}_0^2+q_1^2+q_2^2+q_3^2=1).$ The unit quaternion biggest advantage is not have singularity, but shortcoming has two:

1. redundancy:, all need 4 components to describe this joint no matter rotary joint has several freedom; 2. mould length is 1 constraint.When calculating speed, acceleration, the generalized force of rigid body, use hypercomplex number parametrization joint than better with Eulerian angle with newton-Eulerian equation.Reason is to consider the Eulerian angle singularity problem.But when it come to exercise data is revised, the mould of unit quaternion is about the problem that bundle will cause occurring in the computation optimization constraint, makes and optimizes failure.

Existing rigid multibody dynamics equation mainly grows up from the robot field, describes the joint rotation with Eulerian angle in equation or fully, or describes the joint with hypercomplex number fully and rotate.These two kinds of equations all are not suitable in the correction of human sports trapped data, and reason can not satisfy following two essential condition exactly simultaneously: 1) no singularity; 2) avoid the mistake restricted problem that occurs in the exercise data optimization correction.Therefore need a kind of rigid multibody dynamics equation that can satisfy above two essential condition simultaneously of development.

Because therefore the existence of sum of errors noise in the capturing movement process need be revised the capture movement data, utilize Optimization Model to revise the method that the capture movement data are current a kind of main flows.In the Optimization Model of using, objective function is the gap between exercise data of revising and the original motion data of the catching at present; Constraint function is that the generalized force of six direction on the mass center of human body is zero.This Optimization Model is applicable to the exercise data quality that captures situation preferably, but it is invalid for monumented point is lost in a large number and the monumented point side-play amount is bigger situation.Therefore be necessary to propose a kind of Optimization Model of robust more.

Summary of the invention

An object of the present invention is to overcome existing newton-Eulerian equation and in the human sports trapped data correction, can't satisfy no singularity simultaneously and not have binding defective, thereby proposed a kind of mixing newton-Euler's rigid multibody dynamics equation based on Eulerian angle-hypercomplex number.This equation can solve the problem of Eulerian angle singularity effectively and be that one constraint drops to minimum to the obstruction of optimizing exercise data with quaternary digital-to-analogue length.

Another object of the present invention is higher to the quality requirements of the original motion data of catching in order to overcome traditional Optimization Model, the shortcoming that the constraint condition of optimizing is not easy to be met, thereby a kind of good convergence that not only has is provided, can also guarantees that revised exercise data satisfies the Optimization Model of physical constraint simultaneously.

To achieve these goals, the invention provides a kind of method that noise in the human sports trapped data and error are revised, may further comprise the steps:

10) import human body topological structure file and limbs geometric model file, on computers, set up human geometry's model according to human body topological structure and limbs geometric model then;

20) import the human body gross mass, on computers,,, calculate the body mass distribution parameter of each limbs according to the statistical method in the biomechanics according to the human body limb geometric parameter in the limbs geometric model that obtains in human body gross mass and the step 10); Described body mass distribution parameter comprises the quality of limbs and one 3 * 3 moment of inertia matrix;

30), will adopt in the human sports trapped data input computing machine that existing human motion capture technology obtains, and the movable joint of described human sports trapped data and human body is mated;

40), according to step 30) human sports trapped data that obtains, parametrization human motion joint, when the parametrization joint, at first use Eulerian angle parametrization rotary joint, to each rotary joint, judge whether its Eulerian angle have singularity, then for rotary joint with singularity, Eulerian angle are converted into hypercomplex number, and with this joint of hypercomplex number parametrization;

50), foundation is based on mixing newton-Euler's rigid multibody dynamics equation of Eulerian angle-hypercomplex number, according to step 40) in the parametrization type in the joint that obtains, calculate the broad sense rotation matrix of the relative father's rigid body of current rigid body, the speed of rigid body, the acceleration of rigid body respectively, and utilize the suffered generalized force in joint on the acceleration calculation rigid body of speed, rigid body of broad sense rotation matrix, the rigid body of the relative father's rigid body of current rigid body;

60), foundation is based on the Optimization Model of physical constraint, comprise an objective function and a constraint function in this Optimization Model, described objective function is the integration of quadratic sum in whole motion process of the generalized force of six direction on the human body root rigid body, and described constraint function is that gap between the revised exercise data and the original motion data of catching is less than given threshold values;

70), in step 60) in the Optimization Model set up, the iterations of computation optimization is set, the bound of each optimization variable;

80), according to step 70) in the bound of set iterations and each optimization variable, to step 60) in the Optimization Model set up be optimized, obtain revised motion capture data;

90), revised motion capture data is write output data file by the form that reads in, the intermediate result that target function value, constraint function value and per step iteration are obtained writes in the optimizing process log file simultaneously.

In the technique scheme, in described step 10), described human body topological structure has been described the set membership between human body limb, in described limbs geometric model file, has described the geometric parameter of human body limb, comprises the length parameter of limbs.

In the technique scheme, in described step 10), described human geometry's model can adopt " Hanavan manikin " or " Zatsiorsky manikin " or " Chinese visible human model ".

In the technique scheme, in described step 40) in, describedly judge when whether Eulerian angle have singularity that if the variation of adjacent two frame Eulerian angle is greater than 90 degree, then Eulerian angle have singularity.

In the technique scheme, finish described step 40) after, all Eulerian angle and hypercomplex number are done the B spline-fitting.

In the technique scheme, described step 50) specific implementation step comprises:

Step 5-1), in manikin, begin to do forward recursive from the root rigid body and calculate to the leaf rigid body;

Step 5-2), on each rigid body, judge the parametrization type of current rigid body upper joint, if Eulerian angle, execution in step 5-3), if hypercomplex number, execution in step 5-4);

Step 5-3), to call in the newton-euler dynamical equations of mixing with Eulerian angle be the equation of variable, calculates the broad sense rotation matrix of the relative father's rigid body of current rigid body, the speed of rigid body and the acceleration of rigid body, execution in step 5-5 then);

Step 5-4), calling in the newton-euler dynamical equations of mixing with the hypercomplex number is the equation of variable, the articulation center that calculates the derivative of the rotation matrix of the broad sense rotation matrix of the relative father's rigid body of current rigid body, the rotation matrix of the relative father's rigid body of current rigid body, the relative father's rigid body of current rigid body, the rotation matrix of the relative world coordinate system of current rigid body and the derivative of its inverse matrix, current rigid body to the articulation center of the vector of world coordinates initial point, current rigid body to the derivative of the vector of world coordinates initial point, the speed and the acceleration of rigid body, execution in step 5-5 then);

Step 5-5), do the backward recursive computing by the leaf rigid body to the root rigid body, according to step 5-3) or step 5-4) broad sense rotation matrix, the speed of rigid body and the acceleration of rigid body of the relative father's rigid body of current rigid body that obtains, call in the newton-euler dynamical equations of mixing and calculate the generalized force that is subjected on each joint of Equation for Calculating of generalized force.

The invention has the advantages that: a kind of mixing newton based on Eulerian angle-hypercomplex number-Euler's rigid multibody dynamics equation that is suitable for the motion capture data correction at first is provided.The problem that this equation can not only solve the Eulerian angle singularity effectively also makes quaternary digital-to-analogue length be one constraint to drop to minimum to the obstacle of optimizing exercise data and causing.Next has proposed a kind of new Optimization Model, and this Optimization Model is low to the quality requirements of initial capture movement, and convergence is good, can also guarantee that revised exercise data is in close proximity to the original data of catching satisfying under the situation of physical constraint.The present invention comprehensively uses rigid multibody dynamics, biomechanics, computation optimization technology, and the method for proposition is applicable to human body topological structure and complicated exercise data arbitrarily, has good versatility and practicality.

Description of drawings

Fig. 1 is the process flow diagram of the method that the noise in the human sports trapped data and error are revised of the present invention;

Fig. 2 A utilizes original motion to catch the figure as a result of the human body root rigid body directions X resultant moment that data and revised motion capture data calculate;

Fig. 2 B utilizes original motion to catch the figure as a result of the human body root rigid body Y direction resultant moment that data and revised motion capture data calculate;

Fig. 2 C utilizes original motion to catch the figure as a result of the human body root rigid body Z direction resultant moment that data and revised motion capture data calculate;

Fig. 3 A utilizes original motion to catch the figure as a result that human body root rigid body directions X that data and revised motion capture data calculate is made a concerted effort;

Fig. 3 B utilizes original motion to catch the figure as a result that human body root rigid body Y direction that data and revised motion capture data calculate is made a concerted effort;

Fig. 3 C utilizes original motion to catch the figure as a result that human body root rigid body Z direction that data and revised motion capture data calculate is made a concerted effort;

Fig. 4 is the synoptic diagram of manikin.

The drawing explanation

---the result who utilizes raw data to calculate

The result who utilizes revised data computation to obtain

Embodiment

Below in conjunction with the drawings and specific embodiments method of the present invention is described further.

In one embodiment, the method that the noise in the human sports trapped data and error are revised of the present invention is 2.8GHz a CPU frequency, in save as on the computing machine of 1G and realize that this computing machine adopts WindowsXP operating system.

On the basis of above-mentioned platform, as shown in Figure 1, the method that the noise in the human sports trapped data and error are revised of the present invention may further comprise the steps:

Step 10, input human body topological structure and limbs geometric model file are set up human geometry's model according to human body topological structure and limbs geometric model file.Wherein, described human body topological structure has been described the set membership between human body limb.As shown in Figure 4, in a manikin, the elbow joint on the human body left hand is carpal father joint, and shoulder joint is the father joint of elbow joint.In described limbs geometric model file, then include the geometric parameter of human body limb, as the parameters such as length and width of limbs.Described human body topological structure can obtain from the document of body biomechanics, and the data of described limbs geometric model can measure with tape measure.In this step, the foundation of human geometry's model is ripe prior art, and manikin can adopt " Hanavan manikin ", " Zatsiorsky manikin " and " Chinese visible human model ".The visible list of references 1 of correlation technique: modern sport biomechanics, Zheng Xiuyuan etc., National Defense Industry Press, 2002.

Step 20, input human body gross mass according to the geometric parameter of the human body limb that obtains in human body gross mass and the step 10, with reference to the statistical method in the biomechanics, calculate the body mass distribution parameter of each limbs.Wherein, described mass distribution parameter comprises the quality of limbs (in the rigid multibody dynamics equation, limbs are regarded as rigid body) and 3 * 3 moment of inertia matrix.

Step 30, input human sports trapped data, and motion capture data and movable joint mated.Described motion capture data comprised in the different time, three direction shift value of the rotational angle value of each limbs and human body integral.In this step, described motion capture data and movable joint mate the motion capture data that is meant a certain joint and should interrelate with this joint, rather than interrelate with other joint.For example, carpal motion capture data and wrist joint interrelate, and the motion capture data and the shoulder joint of shoulder joint interrelate.In the present embodiment, the human sports trapped data of input adopts self-defining form VHD, and the file by this form can mate motion capture data and movable joint.

The form of described self-defined VHD is as follows:

First joint title sampling time 1 exercise data, 1 sampling times 2 exercise data 2

Second joint title sampling time 1 exercise data, 1 sampling times 2 exercise data 2

The 3rd joint title sampling time 1 exercise data, 1 sampling times 2 exercise data 2

…………

N joint title sampling time 1 exercise data, 1 sampling times 2 exercise data 2

Step 40, according in the step 30 input human sports trapped data, parametrization human motion joint.Its

Be implemented as follows:

Step 401, usefulness Eulerian angle parametrization rotary joint; In manikin, a joint on the root rigid body is the translation joint, and other joints of manikin all are rotary joint.

Step 402, to each rotary joint, judge that whether the Euler angle of adjacent two frames changes greater than 90 degree, if produced singularity greater than the bright Eulerian angle of 90 kilsyth basalts, execution in step 403, otherwise execution in step 405.With the joint with three degree of freedom is example, and concrete rule is as follows: establish α _i, β _i, γ _iBe three Eulerian angle of certain joint at the i frame.α _I+1, β _I+1, γ _I+1Be the Eulerian angle of i+1 frame.If | α _I+1-α _i|, | β _I+1-β _i|, | γ _I+1-γ _i| some values are arranged greater than 90 degree, then the Eulerian angle in this joint have produced singularity.

Step 403, Eulerian angle are converted into hypercomplex number; It is prior art that Eulerian angle transform hypercomplex number, the visible list of references 2 of its specific implementation: calculate dynamics of multibody systems, Hong Jiazhen, Science Press 2002.

Step 404, usefulness hypercomplex number parametrization rotary joint;

Step 405, all Eulerian angle and hypercomplex number are carried out the B spline-fitting, to reduce the number of independent variable.

Step 50, set up mixing newton-Euler's rigid multibody dynamics equation based on Eulerian angle-hypercomplex number; This equation is the equation of one group of recurrence.According to the joint parameter type, dynamically call the corresponding calculated equation.It is implemented as follows.

Step 501, in manikin, begin to do forward recursive from the root rigid body and calculate to the leaf rigid body, in each rigid body, finish following operation.

The parametrization type of step 5011, the current rigid body upper joint of judgement, if Eulerian angle, execution in step 5012, if hypercomplex number, execution in step 5013.

Step 5012, to call in the newton-euler dynamical equations of mixing with Eulerian angle be below the Equation for Calculating of variable:

1), the broad sense rotation matrix of the relative father's rigid body of current rigid body, shown in equation (1);

$X_i^j=\left[\begin{array}{cc}{R}_i^j& 0\\ -R_i^j(l_i^{\&prime;}\&times;)& R_i^j\end{array}\right]---\left(1\right)$

R _i ^j: the rotation matrix of coordinate system i relative coordinate system j;

l _i': the coordinate figure of the initial point of coordinate system j in (rotation before) coordinate system i;

Wherein symbol " * " is defined as follows

$\left[\begin{array}{c}x\\ y\\ z\end{array}\right]\&times;=\left[\begin{array}{ccc}0& -z& y\\ z& 0& -x\\ -\mathrm{<figure-callout\; id="0"\; label="y\; x"\; filenames="A20061011307100161.png,A20061011307100162.png"\; state="\{\{state\}\}">y</figure-callout>}& x& 0\end{array}\right]$

Especially, for the root rigid body, itself does not have father's rigid body, can only do translation motion, the R in its broad sense rotation matrix _i ^jBe 3 to take advantage of 3 unit matrix.

2), calculate the speed and the acceleration of rigid body.When the speed of use Eulerian angle calculating rigid body and acceleration, accounting equation is shown in equation (2) and equation (3):

$v_i^{\&prime;}=X_{i-1}^i{v}_{i-1}^{\&prime;}+s_i^{\&prime;}{\stackrel{\&CenterDot;}{\mathrm{\&theta;}}}_i---\left(2\right)$

Wherein, v _iThe space velocity of i rigid body of ' expression; a _iThe steric acceleration of i rigid body of ' expression; X _i ^jDenotation coordination is the broad sense rotation matrix of i to coordinate system j; s _iThe rotating shaft in i joint of ' expression; θ _iRepresent the Euler angle value for rotary joint, represent shift value for the translation joint.

Single order, second derivative that expression is corresponding.

Expression becomes one 6 * 1 vector into 6 * 6 matrix, and its rule is as follows:

Step 5013, to call in the newton-euler dynamical equations of mixing with the hypercomplex number be below the Equation for Calculating of variable:

1), the broad sense rotation matrix X of the relative father's rigid body of current rigid body _i ^j, can be with reference to equation (1) to the calculating of space selection matrix;

2), the rotation matrix R of the relative father's rigid body of current rigid body _i ^jTo rotation matrix R _i ^jThe explanation of calculating behind equation (5) in concrete description is arranged.

3), the derivative of the rotation matrix of the relative father's rigid body of current rigid body The derivative of described rotation matrix By to rotation matrix R _i ^jDifferentiate obtains;

4), the rotation matrix R of the relative world coordinate system of current rigid body _i ⁰And the derivative of its inverse matrix

5), the articulation center of current rigid body is to the vector r of world coordinates initial point _i';

6), the articulation center of current rigid body is to the derivative of the vector of world coordinates initial point

7), computing velocity, acceleration.When utilizing hypercomplex number to calculate the speed of rigid body and acceleration, accounting equation is shown in equation (4) and equation (5):

$v_i^{\&prime;}=X_{i-1}^i{v}_{i-1}^{\&prime;}+\left[\begin{array}{c}{\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}_i^{\&prime;}\\ 0\end{array}\right]---\left(4\right)$

$a_i^{\&prime;}=X_{i-1}^i{a}_{i-1}^{\&prime;}+\left[\begin{array}{c}{R}_0^i{\stackrel{\&CenterDot;}{R}}_i^0{\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}_i^{\&prime;}+{\stackrel{\&CenterDot;}{\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}}_i^{\&prime;}\\ {\stackrel{\&CenterDot;}{r}}_i^{\&prime;}\&times;{\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}_i^{\&prime;}\end{array}\right]---\left(5\right)$

Wherein, $R_i^j=\left[\begin{array}{ccc}2q_{i,0}^2-1+2q_{i,1}^2& 2q_{i,1}{q}_{i,2}-2q_{i,0}{q}_{i,3}& 2q_{i,1}{q}_{i,3}+2q_{i,0}{q}_{i,2}\\ 2q_{i,1}{q}_{i,2}+2q_{i,0}{q}_{i,3}& 2q_{i,0}^2-1+2q_{i,2}^2& 2q_{i,2}{q}_{i,3}-2q_{i.0}{q}_{i,1}\\ 2q_{i,1}{q}_{i,3}-2q_{i,0}{q}_{i,2}& 2q_{i,2}{q}_{i,3}+2q_{i,0}{q}_{i,1}& 2q_{i,0}^2-1+2q_{i,3}^2\end{array}\right]$

$R_0^i=R_{i-1}^i{R}_0^{i-1};{\stackrel{\&CenterDot;}{R}}_0^i={\stackrel{\&CenterDot;}{R}}_{i-1}^i{R}_0^{i-1}+R_{i-1}^i{\stackrel{\&CenterDot;}{R}}_0^{i-1};$

${\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}_i^{\&prime;}=2L_i{\stackrel{\&CenterDot;}{q}}_i;{\stackrel{\&CenterDot;}{\stackrel{\&RightArrow;}{\mathrm{\&omega;}}}}_i^{\&prime;}=2L_i{\stackrel{\&CenterDot;\&CenterDot;}{q}}_i$

$L_i=\left[\begin{array}{cccc}-q_{i,1}& q_{i,0}& q_{i,3}& -q_{i,2}\\ -q_{i,2}& -q_{i,3}& q_{i,0}& q_{i,1}\\ -q_{i,3}& q_{i,2}& -q_{i,1}& q_{i,0}\end{array}\right]$

R _i ^jDenotation coordination is the rotation matrix of i to coordinate system j; v _iThe space velocity of i rigid body of ' expression; a _iThe steric acceleration of i rigid body of ' expression; X _i ^jDenotation coordination is the broad sense rotation matrix of i to coordinate system j; q _iRepresent the hypercomplex number that rotate in i joint; ω _iThe angular velocity of i rigid body of ' expression; Represent the angular velocity of i rigid body with respect to i-1 rigid body; l _iThe initial point of i-1 coordinate system of ' expression is to the vector of i coordinate origin; r _iI coordinate origin of ' expression is to the vector of world coordinate system.

In above-mentioned variable, the top of a variable has the first order derivative that a point is represented this variable, has the second derivative of this variable of expression of two points.The upper right corner has an expression change amount of casting aside and is arranged in local coordinate system,

Step 502, do the backward recursive computing to the root rigid body, in each rigid body, set up newton-euler dynamical equations, utilize the generalized force that is subjected on this Equation for Calculating joint by the leaf rigid body.The accounting equation of this generalized force is shown in equation (6).

$I_i^{\&prime;}=\left[\begin{array}{cc}-m_i\underset{\&OverBar;}{c_i}\&times;& m_i\\ \underset{\&OverBar;}{I_i}-m_i\underset{\&OverBar;}{c_i}\&times;c_i\&times;& m_i\underset{\&OverBar;}{c_i}\&times;\end{array}\right]$

m _i: the quality of i root rigid body;

I _i: i root rigid body is with respect to the moment of inertia matrix of barycenter;

c _i: the coordinate of barycenter in i frame of i root rigid body;

Expression becomes one 6 * 1 vector into 6 * 6 matrix, in equation (3) conversion process is had corresponding explanation.

Calculate in the process of generalized force at newton-euler dynamical equations that above-mentioned utilization mixes, for the joint of representing with Eulerian angle, when calculating generalized force, speed and acceleration that employing is calculated by Eulerian angle, for the joint of representing with hypercomplex number, when calculating generalized force, adopt the speed and the acceleration that obtain by hypercomplex number.

Step 60, set up Optimization Model based on physical constraint.The foundation of this Optimization Model comprises following steps:

Step 601, set up objective function, described objective function is the integration of quadratic sum in whole motion process of the generalized force of six direction on the human body root rigid body, and the expression formula of this function is shown in equation (7):

${\&Integral;}_{t=t_0}^{t=t_1}\underset{i=1}{\overset{6}{\mathrm{\&Sigma;}}}{f}_i^2\left(t\right)---\left(7\right)$

Step 602, set up constraint function, described constraint function is that gap between the revised exercise data and the original motion data of catching is less than given threshold values.

|x _i(t _j)-x _i，j|＜H (8)

Wherein, x _i(t _j) represent that i joint is at t _jExercise data constantly, x _{I, j}The original motion data that expression is corresponding, H represents given threshold values.

Step 603, the iterations of computation optimization is set, the bound of each optimization variable.Optimization variable described in this step is meant the reference mark of the B batten of describing joint motions.Therefore can not guarantee to find globally optimal solution because all practical optimized Algorithm all can only find locally optimal solution, an iterations need be set make and reasonably finding approximate locally optimal solution in the time.The bound that optimization variable is set is in order to reduce the field of definition of variable, to accelerate to find the speed of local optimum.

Step 70, use progressively QUADRATIC PROGRAMMING METHOD FOR solving-optimizing problem.Progressively QUADRATIC PROGRAMMING METHOD FOR is a prior art, utilizes the progressively visible list of references 3 of specific implementation of QUADRATIC PROGRAMMING METHOD FOR solving-optimizing problem: Optimum Theory and method, Yuan Yaxiang, Sun Wenyu, Science Press 1999.

Step 80, satisfy the physical constraint human body movement data after being optimized, the correction motion that calculates is caught data write output data file by the form that reads in, the intermediate result that target function value, constraint function value and per step iteration are obtained writes an optimizing process log file simultaneously.

Utilize the method that the noise in the human sports trapped data and error are revised of the present invention, can eliminate the sum of errors noise that exists in the exercise data of being caught well by the human motion capture technology.Fig. 2 is among the embodiment, utilizes original motion to catch the figure as a result that data and revised motion capture data calculate suffered resultant moment on the human body root rigid body respectively.Fig. 2 A represents the resultant moment of directions X, Fig. 2 B represents the resultant moment of Y direction, Fig. 2 C represents the resultant moment of Z direction, wherein, in Fig. 2 A, Fig. 2 B, Fig. 2 C, fine rule represents to be it is calculated that by the original motion catches value of the resultant moment that obtains, and thick line is represented the value of the resultant moment that calculated by revised motion capture data, as can be seen from the figure, by revise resultant moment that the back exercise data calculates than the resultant moment that calculates by former book exercise data more near theoretical value 0.Fig. 3 utilizes original motion to catch data and revised motion capture data calculates the suffered figure as a result that makes a concerted effort on the human body root rigid body respectively, Fig. 3 A represents making a concerted effort of directions X, Fig. 3 B represents making a concerted effort of Y direction, Fig. 3 C represents making a concerted effort of Z direction, wherein, in Fig. 3 A, Fig. 3 B, Fig. 3 C, fine rule represents to it is calculated that the value of making a concerted effort that obtains by the original motion catches, and thick line is represented the value of making a concerted effort that calculated by revised motion capture data.As can be seen from the figure, by making a concerted effort that exercise data after revising calculates than making a concerted effort of calculating by former book exercise data more near theoretical value 0.Fig. 2 and Fig. 3 have obviously reduced error and interference of noise after having illustrated and having utilized the inventive method that exercise data is revised.

Claims (6)

1, a kind of method that noise in the human sports trapped data and error are revised may further comprise the steps:

10) import human body topological structure file and limbs geometric model file, on computers, set up human geometry's model according to human body topological structure and limbs geometric model then;

20) import the human body gross mass, on computers,,, calculate the body mass distribution parameter of each limbs according to the statistical method in the biomechanics according to the human body limb geometric parameter in the limbs geometric model that obtains in human body gross mass and the step 10); Described body mass distribution parameter comprises the quality of limbs and one 3 * 3 moment of inertia matrix;

30), will adopt in the human sports trapped data input computing machine that existing human motion capture technology obtains, and the movable joint of described human sports trapped data and human body is mated;

40), according to step 30) human sports trapped data that obtains, parametrization human motion joint, when the parametrization joint, at first use Eulerian angle parametrization rotary joint, to each rotary joint, judge whether its Eulerian angle have singularity, then for rotary joint with singularity, Eulerian angle are converted into hypercomplex number, and with this joint of hypercomplex number parametrization;

50), foundation is based on mixing newton-Euler's rigid multibody dynamics equation of Eulerian angle-hypercomplex number, according to step 40) in the parametrization type in the joint that obtains, calculate the broad sense rotation matrix of the relative father's rigid body of current rigid body, the speed of rigid body, the acceleration of rigid body respectively, and utilize the suffered generalized force in joint on the acceleration calculation rigid body of speed, rigid body of broad sense rotation matrix, the rigid body of the relative father's rigid body of current rigid body;

60), foundation is based on the Optimization Model of physical constraint, comprise an objective function and a constraint function in this Optimization Model, described objective function is the integration of quadratic sum in whole motion process of the generalized force of six direction on the human body root rigid body, and described constraint function is that gap between the revised exercise data and the original motion data of catching is less than given threshold values;

70), in step 60) in the Optimization Model set up, the iterations of computation optimization is set, the bound of each optimization variable;

80), according to step 70) in the bound of set iterations and each optimization variable, to step 60) in the Optimization Model set up be optimized, obtain revised motion capture data;

90), revised motion capture data is write output data file by the form that reads in, the intermediate result that target function value, constraint function value and per step iteration are obtained writes in the optimizing process log file simultaneously.

2, the method that the noise in the human sports trapped data and error are revised according to claim 1, it is characterized in that, in described step 10), described human body topological structure has been described the set membership between human body limb, in described limbs geometric model file, describe the geometric parameter of human body limb, comprised the length parameter of limbs.

3, the method that the noise in the human sports trapped data and error are revised according to claim 1, it is characterized in that, in described step 10), described human geometry's model can adopt " Hanavan manikin " or " Zatsiorsky manikin " or " Chinese visible human model ".

4, the method that the noise in the human sports trapped data and error are revised according to claim 1, it is characterized in that, in described step 40) in, describedly judge when whether Eulerian angle have singularity, if the variation of adjacent two frame Eulerian angle is greater than 90 degree, then Eulerian angle have singularity.

5, the method that the noise in the human sports trapped data and error are revised according to claim 1 is characterized in that, finishes described step 40) after, all Eulerian angle and hypercomplex number are done the B spline-fitting.

6, the method that the noise in the human sports trapped data and error are revised according to claim 1 is characterized in that described step 50) the specific implementation step comprise:

Step 5-1), in manikin, begin to do forward recursive from the root rigid body and calculate to the leaf rigid body;

Step 5-2), on each rigid body, judge the parametrization type of current rigid body upper joint, if Eulerian angle, execution in step 5-3), if hypercomplex number, execution in step 5-4);

Step 5-3), to call in the newton-euler dynamical equations of mixing with Eulerian angle be the equation of variable, calculates the broad sense rotation matrix of the relative father's rigid body of current rigid body, the speed of rigid body and the acceleration of rigid body, execution in step 5-5 then);

Step 5-4), calling in the newton-euler dynamical equations of mixing with the hypercomplex number is the equation of variable, the articulation center that calculates the derivative of the rotation matrix of the broad sense rotation matrix of the relative father's rigid body of current rigid body, the rotation matrix of the relative father's rigid body of current rigid body, the relative father's rigid body of current rigid body, the rotation matrix of the relative world coordinate system of current rigid body and the derivative of its inverse matrix, current rigid body to the articulation center of the vector of world coordinates initial point, current rigid body to the derivative of the vector of world coordinates initial point, the speed and the acceleration of rigid body, execution in step 5-5 then);

Step 5-5), do the backward recursive computing by the leaf rigid body to the root rigid body, according to step 5-3) or step 5-4) broad sense rotation matrix, the speed of rigid body and the acceleration of rigid body of the relative father's rigid body of current rigid body that obtains, call in the newton-euler dynamical equations of mixing and calculate the generalized force that is subjected on each joint of Equation for Calculating of generalized force.

Images