CN101515371B - Human body movement data fragment extracting method - Google Patents

Abstract

The invention discloses a human body movement data fragment extracting method. The method adopts the computer movement data compiling and the human body model building knowledge to realize the automatic intelligent identification and split of the action fragment with the basic semantic unit in the movement data. At first, the movement capture data needing to extract the action fragment is inputted, the kinematics model of human body is renewed, the centroidal speed is calculated out based on the physical knowledge of each joint and the instant speed, the centroidal speed of corresponding frame in the data sequence is in the spectrum analysis and split based on the obtaining of the centroidal speed, the action fragment with the basic semantic unit is extracted. In the method, because the centroidal speed is calculated out and the speed is in clustering extremume split, the accuracy and the efficiency of the action identification are improved, the low efficiency and the subjectivity of handwork extracting, the efficiency for the reuse of the movement capture data in the human body part animation producing process.

Description

Human body movement data fragment extracting method

Technical field

The present invention relates to a kind of extraction method of human action fragment, be specifically related to a kind of gaining knowledge and handle the semantic cutting method of the original motion data that obtains by the capturing movement equipment with the exercise data editing technique, belong to the general field of computing machine motion capture data editor and human body role animation based on human motion.

Background technology

Along with the development of exercise data reuse technology, the motion segmentation technology becomes the important research focus in human body role animation field gradually.Although still be in the elementary stage about this Study on Technology at present, a lot of relatively effective methods have been arranged.

" the Motion Abstraction and Mapping withSpatial Constraints " that delivered on meeting CAPTECH in 1998 discloses a kind of zero point that detects movable joint momentum acceleration of utilizing and judged whether the human body role carries out mutual method with the external world." the Automated Derivation of Primitives for Movement Classification " that delivered on periodical Autonomous Robots in 2002 utilizes this method to come the motion sequence data are carried out segmentation, and the result of segmentation is carried out based on the classification of principal component analysis (PCA) algorithm (PCA) with based on the cluster of K-Means algorithm.But this algorithm based on the acceleration zero point only can be effective at the eurythmy pattern, can not be by widespread in all types of motions.

In recent years, the motion segmentation technology begins to be conceived to for each human joint points analytically." Motion Modeling for On-Line Locomotion Synthesis " as delivering on SCA in 2005 discloses a kind of human motion dividing method based on centroid trajectory (COM)." the Motion Synthesis from Annotations " that delivered on SIGGRAPH in 2003 discloses a kind of method of utilizing support vector machine (SVM) sorter to come interactively mark exercise data.Although this automanual sorter can be applicable to the mark of motion preferably, this mark framework based on frame can't be caught a conversion process between the variety classes action very accurately.The represented human motion posture of single frame is because the shortcoming on the granularity, can't be really well well mark the whole flow process of motion." the Efficient Content-Based Retrieval of Motion Capture Data " that on periodical TOG, delivered in 2005, disclose a kind of method of motion search, this method is searched for by differentiating the motion feature of representing with Boolean function.When time shaft was out of shape, this method was caught the motion fragment that functional value changes.Same method also has been used on " the Motion Modeling for On-Line LocomotionSynthesis " that was published in 2005 on the SCA, wherein represent system by designing a kind of motion fragment based on symbol, this method has realized rule-based sorting technique.But the part that this method can not represent in the motion segmentation is empirical, and the effect of cutting apart exists stiff problem to a great extent.

Summary of the invention

The objective of the invention is provides a kind of extraction method based on the systemic velocity extreme value in order to overcome among the present exercise data editor for the complicacy and the subjectivity of human motion fragment manual extraction.

Human body movement data fragment extracting method comprises the steps:

1) utilizes people's kinematics model and exercise data feature, based on each articulation point physical knowledge and instantaneous velocity, the method for designing and calculating systemic velocity;

2) based on the systemic velocity computing method of step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction.

Described people's kinematics model and the exercise data feature utilized, based on each articulation point physical knowledge and instantaneous velocity, the method step of designing and calculating systemic velocity:

(a) motion capture data of input standard format is chosen m in the manikin articulation point the most representative as candidate's articulation point; Length in input is the motion capture data sequence S={f of n+1 ₀, f ₁..., f _nIn, f wherein _iHuman body movement data by the corresponding frame of frame index i; Candidate's articulation point information vector f of human body in every frame _i={ P _I0, V _I0, P _I1, V _I1..., P _{I (m-1)}, V _{I (m-1)}Represent P wherein _IjAnd V _IjRepresent that respectively candidate's articulation point j is at frame f _iPairing spatial positional information of timestamp i and velocity information; Its medium velocity V _IjAccount form as follows:

$V_{\mathrm{ij}}=\frac{\left|\right|P_{\mathrm{ij}}-P_{(i-1)j}\left|\right|+\left|\right|P_{(i+1)j}-P_{\mathrm{ij}}\left|\right|}{2}\×\mathrm{fps}---1$

Wherein ‖ ‖ represents Euler's range formula, and fps is the frame per second of motion capture data; Obtained each frame f among the frame sequence S _iCandidate's articulation point information P of middle human body _iAnd V _iAfter, calculate barycenter at each frame f _iPairing systemic velocity V _I-COMInformation:

$V_{i-\mathrm{COM}}=\underset{j=0}{\overset{m}{\mathrm{\Σ}}}{w}_j{V}_{\mathrm{ij}}---2$

W wherein _jBe V _IjCorresponding weighted value, according to different articulation points for the percentage contribution of systemic velocity and difference, acquisition human sports trapped data sequence { f ₀, f ₁..., f _nPairing continuous systemic velocity set { V _0-COM, V _1-COM..., V _N-COM.

Described systemic velocity computing method based on step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction step:

(b) obtain the pairing continuous systemic velocity set { V of human sports trapped data sequence _0-COM, V _1-COM..., V _N-COMAfter, sets of speeds is carried out traversal one time, obtain speed V maximum in the set _MaxWith minimum speed V _Min, the threshold value θ of a filtration is set _Thre, computing formula is as follows:

${\mathrm{\θ}}_{\mathrm{thre}}=\frac{V_{\max }-V_{\min }}{\mathrm{slack}}---3$

Wherein slack is adjustable division parameter, and the difference between minimum and maximum is divided between several dividing regions, will satisfy { V in the systemic velocity set _i≤ V _Min+ θ _Thre| V _i∈ { V _ThrePut into and divide set { V _ThreIn;

(c) for dividing set { V _ThreIn each systemic velocity V _I-COM, drop on (θ between the index area for all index value i _Range,+θ _Range) in speed put into one and divide to assemble subclass { V _RangeIn, θ wherein _RangeIt is adjustable radius parameter; Choose the pairing concentration range threshold value of index, obtain for dividing set { V _ThreSeveral divide to assemble the intersection { { V of subclass _1-range, { V _2-range..., { V _K-range, wherein k is by being obtained to assemble number; Ask for then to divide and assemble subclass { V _I-rangeCorresponding central speed V _I-center:

$V_{i-\mathrm{center}}=\frac{\underset{i=1}{\overset{\mathrm{\λ}}{\mathrm{\Σ}}}{w}_i{V}_{i-\mathrm{COM}}}{\mathrm{\λ}},$ $i-\mathrm{center}=i+\frac{{\mathrm{\θ}}_{\mathrm{range}}}{2}---4$

V wherein _I-centerAssemble subclass { V for dividing _I-rangeIn all speed V _I-COMWeighted mean, w _iBe the weighted value of speed correspondence, index i-center is the intermediate value between this index area, and λ assembles subclass { V for dividing _I-rangeIn the number of all elements; A complete division set { V who obtains _ThreCorresponding V _CenterSet { V _1-center, V _2-center..., V _K-center, according to the V that obtains _CenterSet, to the index of frame according to 1-center, 2-center ..., k-center} is cut apart exercise data as cut-point, extracts exercise data between two cut-points as the action fragment.

Description of drawings

Fig. 1 is a human skeleton model of the present invention;

Fig. 2 is a systemic velocity frequency spectrum of the present invention.

Embodiment

Human body movement data fragment extracting method comprises the steps:

1) utilizes people's kinematics model and exercise data feature, based on each articulation point physical knowledge and instantaneous velocity, the method for designing and calculating systemic velocity;

2) based on the systemic velocity computing method of step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction.

Described people's kinematics model and the exercise data feature utilized, based on each articulation point physical knowledge and instantaneous velocity, the method step of designing and calculating systemic velocity:

(a) motion capture data of standard format is chosen m in the manikin articulation point the most representative as candidate's articulation point; Length in input is the motion capture data sequence S={f of n+1 ₀, f ₁..., f _nIn, f wherein _iHuman body movement data by the corresponding frame of frame index i; Candidate's articulation point information vector f of human body in every frame _i={ P _I0, V _I0, P _I1, V _I1..., P _{I (m-1)}, V _{I (m-1)}Represent P wherein _IjAnd V _IjRepresent that respectively candidate's articulation point j is at frame f _iPairing spatial positional information of timestamp i and velocity information; Its medium velocity V _IjAccount form as follows:

$V_{\mathrm{ij}}=\frac{\left|\right|P_{\mathrm{ij}}-P_{(i-1)j}\left|\right|+\left|\right|P_{(i+1)j}-P_{\mathrm{ij}}\left|\right|}{2}\×\mathrm{fps}---1$

Wherein || || expression Euler range formula, fps is the frame per second of motion capture data; Obtained each frame f among the frame sequence S _iCandidate's articulation point information P of middle human body _iAnd V _iAfter, calculate barycenter at each frame f _iPairing systemic velocity V _I-COMInformation:

$V_{i-\mathrm{COM}}=\underset{j=0}{\overset{m}{\mathrm{\Σ}}}{w}_j{V}_{\mathrm{ij}}---2$

W wherein _jBe V _IjCorresponding weighted value, according to different articulation points for the percentage contribution of systemic velocity and difference, acquisition human sports trapped data sequence { f ₀, f ₁..., f _nPairing continuous systemic velocity set { V _0-COM, V _1-COM..., V _N-COM.

Described systemic velocity computing method based on step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction step:

(b) obtain the pairing continuous systemic velocity set { V of human sports trapped data sequence _0-COM, V _1-COM..., V _N-COMAfter, sets of speeds is carried out traversal one time, obtain speed V maximum in the set _MaxWith minimum speed V _Min, the threshold value θ of a filtration is set _Thre, computing formula is as follows:

${\mathrm{\θ}}_{\mathrm{thre}}=\frac{V_{\max }-V_{\min }}{\mathrm{slack}}---3$

Wherein slack is adjustable division parameter, and the difference between minimum and maximum is divided between several dividing regions, will satisfy { V in the systemic velocity set _i≤ V _Min+ θ _Thre| V _i∈ { V _ThrePut into and divide set { V _ThreIn;

(c) for dividing set { V _ThreIn each systemic velocity V _I-COM, drop on (θ between the index area for all index value i _Range,+θ _Range) in speed put into one and divide to assemble subclass { V _RangeIn, θ wherein _RangeIt is adjustable radius parameter; Choose the pairing concentration range threshold value of index, obtain for dividing set { V _ThreSeveral divide to assemble the intersection { { V of subclass _1-range, { V _2-range..., { V _K-range, wherein k is by being obtained to assemble number; Ask for then to divide and assemble subclass { V _I-rangeCorresponding central speed V _I-center:

$V_{i-\mathrm{center}}=\frac{\underset{i=1}{\overset{\mathrm{\λ}}{\mathrm{\Σ}}}{w}_i{V}_{i-\mathrm{COM}}}{\mathrm{\λ}},$ $i-\mathrm{center}=i+\frac{{\mathrm{\θ}}_{\mathrm{range}}}{2}---4$

V wherein _I-centerAssemble subclass { V for dividing _I-rangeIn all speed V _I-COMWeighted mean, w _iBe the weighted value of speed correspondence, index i-center is the intermediate value between this index area, and λ assembles subclass { V for dividing _I-rangeIn the number of all elements; A complete division set { V who obtains _ThreCorresponding V _CenterSet { V _1-center, V _2-center..., V _K-center, according to the V that obtains _CenterSet, to the index of frame according to 1-center, 2-center ..., k-center} is cut apart exercise data as cut-point, extracts exercise data between two cut-points as the action fragment.

Embodiment

(1) human sports trapped data of input standard format is imported the file of the human motion standard .trc form that the near-infrared optical capture systems obtained, per second 24 frames in the present embodiment; Choose in the manikin 18 articulation points the most representative as candidate's articulation point, belong to the part that four limbs, cervical vertebra, the several motions of head are concentrated respectively, human skeleton information as shown in Figure 1;

Length in input is the motion capture data sequence S={f of n+1 ₀, f ₁..., f _nIn, f wherein _iExercise data by the corresponding frame of frame index i; Candidate's articulation point information vector f of human body in every frame _i={ P _I0, V _I0, P _I1, V _I1..., P _{I (m-1)}, V _{I (m-1)}Represent P wherein _IjAnd V _IjRepresent that respectively candidate's articulation point j is at frame f _iPairing spatial positional information of timestamp and velocity information, m represents the number of candidate's articulation point, m is 18 in the present embodiment, its medium velocity V _IjAccount form as follows:

$V_{\mathrm{ij}}=\frac{\left|\right|P_{\mathrm{ij}}-P_{(i-1)j}\left|\right|+\left|\right|P_{(i+1)j}-P_{\mathrm{ij}}\left|\right|}{2}\×\mathrm{fps}$

Wherein || || expression Euler range formula, fps is the frame per second of motion capture data, is 24 frame per seconds in the present embodiment; Obtained each frame f among the frame sequence S _iCandidate's articulation point information P of middle human body _iAnd V _iAfter, calculate barycenter at each frame f _iPairing systemic velocity V _I-COMInformation:

$V_{i-\mathrm{COM}}=\underset{j=0}{\overset{m-1}{\mathrm{\Σ}}}{w}_j{V}_{\mathrm{ij}}$

W wherein _jBe V _IjCorresponding weighted value, for the percentage contribution of systemic velocity and difference, in the present embodiment, the weights of candidate's articulation point correspondence that belong to four limbs are bigger according to different articulation points, and the weights that belong to cervical vertebra and head are less; Obtain human sports trapped data sequence { f ₀, f ₁..., f _nPairing continuous systemic velocity set { V _0-COM, V _1-COM..., V _N-COM, as shown in Figure 2.

(2) obtain the pairing continuous systemic velocity set { V of human sports trapped data sequence _0-COM, V _1-COM..., V _N-COMAfter, utilize the bubbling algorithm that sets of speeds is carried out traversal one time in this example, obtain speed V maximum in the set _MaxWith minimum speed V _Min, the threshold value θ of a filtration is set _Thre, computing formula is as follows:

${\mathrm{\θ}}_{\mathrm{thre}}=\frac{V_{\max }-V_{\min }}{\mathrm{slack}}$

Wherein slack is adjustable division parameter, and the difference between minimum and maximum is divided between several dividing regions, will satisfy { V in the systemic velocity set _i≤ V _Min+ θ _Thre| V _i∈ { V _ThrePut into and divide set { V _ThreIn, obtain the sets of speeds that systemic velocity set medium velocity is close to the lowest point value; In the present embodiment, the default setting of slack is that the adjusting slack that 6 whiles can be real-time obtains best effect;

For dividing set { V _ThreIn each systemic velocity V _I-COM, drop on (θ between the index area for all index value i _Range,+θ _Range) in speed put into one and divide to assemble subclass { V _RangeIn, θ wherein _RangeBe adjustable radius parameter, wherein θ _RangeBe adjustable radius parameter, choose the pairing concentration range threshold value of index, index value is carried out a cluster, eliminate the information redundancy of index value, in the present embodiment θ _RangeDefault value be 5 simultaneously also can be real-time adjusting θ _RangeObtain best effect; Choose the pairing concentration range threshold value of index, obtain for dividing set { V _ThreSeveral divide to assemble the intersection { { V of subclass _1-range, { V _2-range..., { V _K-range, wherein k is by being obtained to assemble number; Ask for then to divide and assemble subclass { V _I-rangeCorresponding central speed V _I-center:

$V_{i-\mathrm{center}}=\frac{\underset{i=1}{\overset{\mathrm{\λ}}{\mathrm{\Σ}}}{w}_i{V}_{i-\mathrm{COM}}}{\mathrm{\λ}},$ $i-\mathrm{center}=i+\frac{{\mathrm{\θ}}_{\mathrm{range}}}{2}$

V wherein _I-centerAssemble subclass { V for dividing _I-rangeIn all speed V _I-COMWeighted mean, w _iBe the weighted value of speed correspondence, index i-center is the intermediate value between this index area, and λ assembles subclass { V for dividing _I-rangeIn the number of all elements; A complete division set { V who obtains _ThreCorresponding V _CenterSet { V _1-center, V _2-center..., V _K-center, according to the V that obtains _CenterSet, to the index of frame according to { 1-center, 2-center, ..., k-center} is cut apart exercise data as cut-point, extract between two cut-points exercise data as the action fragment, in the present embodiment, the .trc file is carried out header and specific data are made amendment, obtain .trc file corresponding to the human action fragment.

Claims (1)

1. a human body movement data fragment extracting method is characterized in that comprising the steps:

1) utilizes people's kinematics model and exercise data feature, based on each articulation point physical knowledge and instantaneous velocity, the method for designing and calculating systemic velocity;

2) based on the systemic velocity computing method of step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction;

Described people's kinematics model and the exercise data feature utilized, based on each articulation point physical knowledge and instantaneous velocity, the method step of designing and calculating systemic velocity:

(a) motion capture data of input standard format is chosen m in the manikin articulation point the most representative as candidate's articulation point; Length in input is the motion capture data sequence S={f of n+1 ₀, f ₁..., f _nIn, f wherein _iHuman body movement data by the corresponding frame of frame index i; Candidate's articulation point information vector f of human body in every frame _i={ P _I0, V _I0, P _I1, V _I1..., P _{I (m-1)}, V _{I (m-1)}Represent P wherein _IjAnd V _IjRepresent that respectively candidate's articulation point j is at frame f _iPairing spatial positional information of timestamp i and velocity information; Its medium velocity V _IjAccount form as follows:

$V_{\mathrm{ij}}=\frac{\left|\right|P_{\mathrm{ij}}-P_{(i-1)j}\left|\right|+\left|\right|P_{(i+1)j}-P_{\mathrm{ij}}\left|\right|}{2}\×\mathrm{fps}---1$

‖ wherein. ‖ represents Euler's range formula, and fps is the frame per second of motion capture data; Obtained each frame f among the frame sequence S _iCandidate's articulation point information P of middle human body _iAnd V _iAfter, calculate barycenter at each frame f _iPairing systemic velocity V _I-COMInformation:

$V_{i-\mathrm{COM}}=\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{w}_j{V}_{\mathrm{ij}}---2$

W wherein _jBe V _IjCorresponding weighted value, according to different articulation points for the percentage contribution of systemic velocity and difference, acquisition human sports trapped data sequence { f ₀, f ₁..., f _nPairing continuous systemic velocity set { V _0-COM, V _1-COM..., V _N-COM; Described systemic velocity computing method based on step 1), the corresponding systemic velocity sequence of institute's input motion sequence of data frames is carried out spectrum analysis, and analysis result is cut extraction step:

(b) obtain the pairing continuous systemic velocity set { V of human sports trapped data sequence _0-COM, V _1-COM..., V _N-COMAfter, sets of speeds is carried out traversal one time, obtain speed V maximum in the set _MaxWith minimum speed V _Min, the threshold value θ of a filtration is set _Thre, computing formula is as follows:

${\mathrm{\θ}}_{\mathrm{thre}}=\frac{V_{\max }-V_{\min }}{\mathrm{slack}}---3$

Wherein slack is adjustable division parameter, and the difference between minimum and maximum is divided between several dividing regions, will satisfy { V in the systemic velocity set _i≤ V _Min+ θ _Thre| V _i∈ { V _ThrePut into and divide set { V _ThreIn;

(c) for dividing set { V _ThreIn each systemic velocity V _I-COM, drop on (θ between the index area for all index value i _Range,+θ _Range) in speed put into one and divide to assemble subclass { V _RangeIn, θ wherein _RangeIt is adjustable radius parameter; Choose the pairing concentration range threshold value of index, obtain for dividing set { V _ThreSeveral divide to assemble the intersection { { V of subclass _1-range, { V _2-range..., { V _K-range, wherein k is by being obtained to assemble number; Ask for then to divide and assemble subclass { V _I-rangeCorresponding central speed V _I-center:

$V_{i-\mathrm{center}}=\frac{\underset{i=1}{\overset{\mathrm{\λ}}{\mathrm{\Σ}}}{w}_i{V}_{i-\mathrm{COM}}}{\mathrm{\λ}},i-\mathrm{center}=i+\frac{{\mathrm{\θ}}_{\mathrm{range}}}{2}---4$

V wherein _I-centerAssemble subclass { V for dividing _I-rangeIn all speed V _I-COMWeighted mean, w _iBe the weighted value of speed correspondence, index i-center is the intermediate value between this index area, and λ assembles subclass { V for dividing _I-rangeIn the number of all elements; A complete division set { V who obtains _ThreCorresponding V _CenterSet { V _1-center, V _2-center..., V _K-center, according to the V that obtains _CenterSet, to the index of frame according to 1-center, 2-center ..., k-center} is cut apart exercise data as cut-point, extracts exercise data between two cut-points as the action fragment.

Images