CN101216952B - Dynamic spatiotemporal coupling denoise processing method for data catching of body motion - Google Patents
Dynamic spatiotemporal coupling denoise processing method for data catching of body motion Download PDFInfo
- Publication number
- CN101216952B CN101216952B CN2008100101702A CN200810010170A CN101216952B CN 101216952 B CN101216952 B CN 101216952B CN 2008100101702 A CN2008100101702 A CN 2008100101702A CN 200810010170 A CN200810010170 A CN 200810010170A CN 101216952 B CN101216952 B CN 101216952B
- Authority
- CN
- China
- Prior art keywords
- semantic
- data
- human body
- semantic chunk
- chunk
- 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.)
- Active
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a dynamic space-time coupling processing method for noise and missing data processing in the optical human motion capture data, comprising the establishment of a dynamic pulse noise model of the noise data and the missing data in the originally-gathered motion data based on a standby semantic chunk, which is characterized in that: the invention comprises the following steps of: generating the standby semantic chunk based on a matched rigid structure, establishing a human topology model based on semantic nodes, generating an optimal human body structure based on the spatial semantic constraint, establishing a pulse noise model based on the position of the semantic chunk and the dynamic time sequence based on the attitude attribute, filtering noise data and reconstructing missing data. The processing method of the inventive embodiment is applicable for processing the motion data which meet any feature point arrangement of the local rigidity structure with good practicability and robustness.
Description
Technical field
The present invention relates to the human body motion capture technical field, particularly the disposal route of noise data and missing data in the optics human body motion capture data.
Background technology
In recent years, human body movement data obtains and develops into a kind of important data obtaining means in field such as virtual reality, computer vision, video display cartoon making with reconfiguration technique gradually, be that a plurality of subjects such as mathematics, computer graphics, Flame Image Process, data processing interpenetrate, cross one another new branch of science, expanded computer application field, enriched computer animation, virtual reality area research content, had important theory and be worth.Simultaneously, human body movement data obtains with reconfiguration technique also has actual application value widely, can be applicable to many industries such as industry, agricultural, traffic, culture, education and public health and physical culture, particularly, important function and significance is arranged in the application in fields such as medical science, Sports Scientific Research, instructing exercise training, modern video display, animation and game making.Therefore, improve motion-captured precision, efficient, reduce equipment cost, to promoting the fast development of production of film and TV, Entertainment, animation industry, make up the animation industry chain that supports mutually, promote national Development of Cultural Industry and then drive growth of the national economic to have important effect and profound meaning.
Human body movement data obtains with reconfiguration technique and just is being subjected to increasing attention as the most important motion capture system of obtaining the source of human body movement data, classify by principle of work, mode commonly used at present mainly contains: mechanical type, acoustics formula, electromagnetic type and optical profile type.Compare with other motion-captured mode, optical motion capture has data and obtains advantages such as convenience, sampling precision, frequency height, usable range be wide.The optical motion capture system uses a kind of relatively widely human movement capture system at present, this type systematic is generally based on principle of computer vision, special sign in the moving object or luminous point monitors and tracking is gone forward side by side line data are handled, and the data after the processing generally are the data stream based on a position.Usually, human body optical motion capture system equipment comprises high speed infrared video camera, video frequency collection card, Video Controller, graphics workstation and reflective signature ball, and capture rate generally reaches more than per second 60 frames.The processing procedure of optical motion capture system is divided into three major parts: raw data is obtained, and comprises a plurality of cameras are calibrated, and performing artist's gauge point is set, image acquisition and three-dimensional space data reconstruct; The raw data of gathering is carried out data processing, comprise that noise data is handled and missing data is handled; Be human motion reconstruct at last, promptly use exercise data to drive the virtual human body model.
Because aspects such as motion capture device and many orders of image matching algorithm are intrinsic, cause the three-dimensional space position of the signature point that obtains sum of errors puppet data to occur, be referred to as noise data.For example, noise appears in the time of the overlapping projection of the delay projection of signature point, signature point, camera acquisition image and many orders of the image matching error exercise data that all can cause collecting.On the other hand; in the captured target motion process; the situation that the signature point is blocked or signature point is overlapping occurs through regular meeting, the human body movement data that causes obtaining lacks, and will realize that the motion model reconstruct that meets the human body topological structure must rely on complete, correct exercise data.Therefore, for integrality, accuracy and the continuity of restore data, design noise data and missing data detect and disposal route, data are handled very necessary.
In traditional method, carrying out data processing has two big class methods.One class is from the time series angle, follows the tracks of the track of each signature point, sets up curve model at the unique point track, and carries out the filtering of noise data and the recovery of missing data on the geometric locus basis.This method meets the principal character of movement capturing data, and treatment effeciency is higher, has guaranteed the continuity of data in the process of processing.Yet also have intrinsic defective, promptly this method requires the start frame of data correctly need carry out necessary manual initial work usually.Secondly, situation about overlapping for the signature locus of points in the motion process i.e. distance between two different characteristic gauge points is too small, and mistake can appear in tracking, needs manual correction.In addition, this method has also been ignored the human body restriction relation in the three dimensions.Thereby no matter this method all is difficult to reach gratifying effect on treatment effeciency and precision.Another kind of method is at space angle, based on human cinology and principle of dynamics, sets up human cinology and kinetics equation, and is optimized according to the human body topological structure.The benefit of this method is the physical characteristics in conjunction with human motion, improves the accuracy of data processing, and limitation is that calculated amount is often bigger in the processing procedure, thereby causes treatment effeciency lower, has ignored the effective information on the time series simultaneously.
Summary of the invention
The objective of the invention is to by to above two class methods advantages and circumscribed analysis, the human body topology information that extracts in continuity information of extracting in employing optimization, the coordinated time sequence and the three dimensions improves the quality and the efficient of data processing, design a kind of time-domain constraints condition and spatial domain constraint condition of synthesizing and coordinating, have the data processing method of high treatment efficiency, good robustness.
Technical solution of the present invention is achieved in that
A kind of dynamic space-time coupling denoise processing method that is used for the human body motion capture data, comprise will based on the optics human body motion capture data handling system of C++ and openGL shape library exploitation pack into computing machine and by human body optical motion capture system equipment obtain raw data, to the step of the processing and the human motion reconstruct of raw data, it is characterized in that further comprising the steps of:
(1) noise data and the missing data that exists in the exercise data at acquired original set up the dynamic pulse noise model based on alternative semantic chunk;
(2), generate alternative semantic chunk, and be defined as five kinds of basic match-types, i.e. line segment type structure, triangular structure, free quadrilateral structure, diagonal angle quadrilateral structure and rigidity quadrilateral structure based on the rigid structure matching algorithm;
(3), definition semantic chunk and alternative semantic chunk, and all signature points are divided into 11 semantic chunks according to the organization of human body feature, comprise head, chest, waist, left arm, right arm, left hand, the right hand, left leg, right leg, left foot and right crus of diaphragm, wherein head, chest, waist, left hand, the right hand, left foot and right crus of diaphragm are the rigidity quadrilateral structure; Left arm, right arm, left leg and right leg are the line segment type structure;
(4), definition semantic node and each property parameters thereof, and human body topological structure relation is described based on the human body semantic model that 22 semantic nodes are set up in this definition, the property parameters of semantic node comprises: sequence number, title, degree, level, father node, length, direction and position, and wherein sequence number, title, degree, level and father node are defined as the subordinate relation attribute; Length, direction and location definition are the locus attribute;
(5), according to the semantic constraint condition of three types of human body semantic model structures, i.e. distance restraint, angle restriction and direction constrain, whether the alternative semantic chunk that generates in the determination step (2) meets constraint condition; Return respectively according to result of determination and to be or two states not; For missing data, assignment is error identification value e, utilizes it that missing data is converted into noise data, for providing prerequisite with the subsequent treatment unification for noise reduction process;
(6), generate the most rational organization of human body, promptly the most rational one group of semantic chunk;
(7), set up impulsive noise model based on the dynamic time sequence of semantic chunk position and attitude attribute, each semantic chunk is all explained with position in the three dimensions and attitude, wherein the position of rigidity quadrilateral structure can be by center point P (P
x, P
y, P
z) define, attitude can be by a normal vector N (N
x, N
y, N
z) and the vectorial D (D of sensing
x, D
y, D
z) determine; The line segment type structure can be by center point P (P
x, P
y, P
z) and the vectorial D (D of sensing
x, D
y, D
z) define, each semantic chunk is represented by 6 or 9 parameters; In the time domain scope, the value of corresponding parameters constitutes a time series in all frames, and each semantic chunk can be represented by 6 or 9 sub-time serieses; All use sub-time sequence to represent all semantic chunks, just constructed impulsive noise model based on the dynamic time sequence of semantic chunk position and attitude attribute;
(8), 5 linear smoothing algorithms are applied on the time series that step (7) set up, realize the filtering of noise data and the reconstruction processing of missing data.
Described reconstruction processing is that the value of subsequence on each time point with each semantic chunk after handling oppositely is reconstructed into position and the attitude of semantic chunk in three dimensions, thus the exercise data after obtaining handling.
Compared with prior art, advantage of the present invention is conspicuous, mainly show by optimizing the human body topology information that the continuity information extracted in the coordinated time sequence and three dimensions extracts and improve the quality and the efficient of data processing, and make it obtain good robustness.
Description of drawings
Fig. 1 FB(flow block) of the present invention.
Fig. 2 signature point semantic chunk is divided synoptic diagram.
The alternative semantic chunk sectional drawing of semantic chunk " waist " in Fig. 3 three frame different pieces of informations.
The structural representation of the basic matching relationship of 5 classes that defines among Fig. 4 the present invention.
Fig. 5 rigid structure matching process synoptic diagram.
The spatial relationship of the 22 node semantic models that define among Fig. 6 the present invention and topological structure graph of a relation.
Fig. 7 direction constrain example schematic diagram (a) is differentiated for shoulder point, (b) differentiates for left and right sides pin.
The position and the attitude synoptic diagram of Fig. 8 rigidity quadrilateral semantic chunk.
Center point P (the P of head semantic chunk in Fig. 9 set
x, P
y, P
z) view before the processing of the sequence that generated after state (noise sequence that contains peak value) and the processing.
The sectional drawing that one group of dance movement exercise data of Figure 10 is handled, left part is the signature point data before handling, right part is the result after handling.
The result of partial frame in Figure 11 exercise data, from a left side and right be respectively jump over, rope skipping, dancing 1, dancing 2 and the action walked.
In the drawings, 1, the original motion data obtains and the foundation of initial stage dynamic noise model, 2, the foundation of rigid structure Matching Model, 3, the definition of alternative semantic chunk, 4, the definition of human body semantic model, 5, the foundation of semantic constraint condition, 6, the foundation of the most reasonable organization of human body semantic model, 7, the foundation of the impulsive noise model of dynamic time sequence, 8, the filtering of noise data and the reconstruct of missing data.
Embodiment
As Fig. 1-shown in Figure 11.A kind of dynamic space-time coupled processing method that is used for optics human body motion capture data noise and missing data processing, comprise that raw data obtains, the step of original data processing and human motion reconstruct, be intended to the noise data and the missing data that exist in the exercise data at acquired original, foundation is characterized in that also comprising that based on the dynamic pulse noise model of alternative semantic chunk the alternative semantic chunk based on the rigid structure coupling generates, foundation based on the human body topological structure model of semantic node, the generation of reasonable organization of human body based on the space semantic constraint, impulsive noise modelling based on the dynamic time sequence of semantic chunk position and attitude attribute, and the step of the reconstruct of the filtering of noise data and missing data.In the present invention, for setting up the model of human body topological structure, suppose: (1) organization of human body is the rigidity articulated structure; (2) in the motion capture process, signature point position meets human body rigidity articulated structure relation.The particular content of described each step comprises:
1, the step that generates based on the alternative semantic chunk of rigid structure coupling is comprising two basic definitions.Define 1 semantic chunk: the semantic chunk that is combined into that is fixed on signature point on the same human body rigid structure.As shown in Figure 2, all signature points are divided into 11 semantic chunks: head, chest, waist, left arm, right arm, left hand, the right hand, left leg, right leg, left foot and right crus of diaphragm.Define 2 alternative semantic chunks: in a certain frame the original motion data of gathering, might belong to the combination of the signature point of semantic chunk.Be the alternative semantic chunk of semantic chunk " waist " in the three frame different pieces of informations as shown in Figure 3.Based on above two basic definitions, defined the basic matching relationship of 5 classes: line segment coupling, triangle coupling, free quadrilateral coupling, diagonal angle quadrilateral coupling and rigidity quadrilateral coupling are to adapt to the coupling of different semantic chunks, as shown in Figure 4.At first, the static masterplate data of choosing one group of standard attitude are as the reference of coupling, choose upright, both arms usually and stretch attitude as masterplate, as shown in Figure 2, and masterplate data and exercise data are captured in finishing in the signature point wearing, to guarantee the accuracy of coupling.With rigidity quadrilateral shown in Figure 5 is example, and the step that is generated alternative semantic chunk by semantic chunk is as follows:
Step 1: be in the signature point P on the same semantic chunk P in the delivery version
1, P
2, P
3, P
4
Step 2: calculate every distance between two points d
1, d
2, d
3, d
4, d
5, d
6
Step 3: specification error allowed band ε, to adapt to the rigid body form variations;
Step 4: search and calculating in given a certain frame exercise data, if there is one group of some P
1', P
2', P
3', P
4', satisfy equation (1), think that then P ' is the alternative semantic chunk of semantic chunk P.In a given frame exercise data, the alternative semantic chunk number N of certain semantic chunk
i0 (signature point disappearance) be may be, also may exercise data quality and error allowed band ε be depended primarily on greater than 1 (other similar structures).
|d
i′-d
i|＜εi＝1，2，3，4，5，6 (1)
Wherein, d
i' be and d
iDistance between the corresponding point.
Step 5: circulation step 4, search for all qualified alternative semantic chunks.
In like manner be applicable to other match-types, coupling is applied to all semantic chunks, generate the alternative semantic chunk of all semantic chunks.
In addition, for rigidity quadrilateral coupling, there are three points on the same rigid structure visible and the situation of disappearance another one point can reconstruct in local coordinate system lack point according to masterplate and rigid structural relation.
2. based on the step of the foundation of the human body topological structure model of semantic node, comprise definition 3 and definition 4, wherein: define 3 semantic nodes: describe the minimum unit of human body topological structure, characterize by following property parameters:
Sequence number: node serial number, as shown in Figure 6, designated root node " sequence number " is 1;
Title: node physical label;
Degree: the number of the child node of this node is 3 as the root node degree;
Level: the node number between this node and the root node (comprising this node itself), specifying root node " level " is 0;
Father node: the sequence number of the node of this node institute subordinate.Specifying root node " father node " is 0;
Length: this node is to the distance of its father node.Specifying root node " length " is 0.0cm;
Direction: the vector of unit length that this node father node points to this node.Specifying root node " direction " is n
0=(0.0,0.0,0.0);
The position: the node space position can obtain according to root node " position ", " length " and " direction " calculation of parameter.The initialization of root node " position " palpus;
Wherein " sequence number ", " title ", " degree ", " level ", " father node " define the subordinate relation attribute of meaning, the locus attribute of " length ", " direction " and " position " definition meaning.
Define 4 semantic models: the tree model of forming by all semantic nodes.
Based on 22 node semantic models of above-mentioned definition 3, in order to determine the subordinate relation attribute and the locus attribute of semantic model respectively.
3. based on the step of the generation of reasonable organization of human body of space semantic constraint, wherein construct a series of semantic constraint conditions according to human body physical arrangement signature analysis and semantic model basis, be intended to judge whether alternative semantic chunk satisfies constraint condition, satisfy and return "Yes", otherwise return "No".Its constraint condition is:
(1) distance restraint: judge between the two adjacent alternative semantic chunk centers that whether distance compare with respective standard semantic chunk distance between centers within the error allowed band,, otherwise return "No" if return "Yes".
(2) angle restriction: judge between the two adjacent alternative semantic chunks between angle and respective standard semantic chunk that whether angle compare within the error allowed band,, otherwise return "No" if return "Yes".
(3) direction constrain: judge between the two adjacent alternative semantic chunks between direction relations and respective standard semantic chunk that whether direction is that the normalization vector relation is compared identical, if return "Yes", otherwise returns "No".As shown in Figure 7, (a) be 4 direction determinings relations of right and left shoulders point and chest, vector
With plane P
2, P
1, P
4The normal orientation unanimity is then returned "Yes", expression P
5Be left side shoulder point, otherwise be right shoulder point; (b) be the structure resolution of left and right sides pin, vector
With plane P
1, P
3, P
2, normal orientation is consistent to be left foot, otherwise is right crus of diaphragm.
For setting up unified noise model, be empty situation for alternative semantic chunk, at three kinds of decision conditions, formulate error identification value e respectively, error identification value e be one can be by the obvious value of identification of system.Distance restraint e=1000cm; Angle restriction e=360 °; Direction constrain e=1.1, simultaneously, semantic constraint can be realized the auto-initiation of exercise data, the start frame of exercise data need not be carried out manual demarcation.
Can obtain one group of alternative semantic chunk that satisfies all semantic constraint conditions through semantic constraint, be called the most reasonable organization of human body.
4. based on the step of the impulsive noise modelling of semantic chunk dynamic time sequence, wherein, after through the said method step process,, can correctly identify the semantic chunk in the Rational structure for most of frame data, still, still there are noise and disappearance.For missing data, carry out improper value and indicated processing, therefore missing data is converted into the noise figure of sudden change.On this basis, can construct a kind of noise spike model based on semantic chunk position and attitude.As shown in Figure 8, be example with rigidity quadrilateral semantic chunk, each semantic chunk can be characterized by position in the three dimensions and attitude; Wherein the position can be by center point P (P
x, P
y, P
z) defining (can choosing wherein, a diagonal line mid point substitutes), attitude can be by a normal vector N (N
x, N
y, N
z) and the vectorial D (D of sensing
x, D
y, D
z) determine.Like this, each semantic chunk can be characterized by 9 parameters in each frame; In the time domain scope, the value of corresponding parameters constitutes a time series in all frames, and each semantic chunk can be characterized by 9 time serieses.All use the time sequence to represent all semantic chunks, just constructed impulsive noise model, as shown in Figure 9, be respectively the center point P (P of a certain semantic chunk based on the dynamic time sequence of semantic chunk position and attitude attribute
x, P
y, P
z) noise model that generated.Through actual analysis, therefore this noise model coincidence impulse noise model feature can use the impulsive noise disposal route to handle.In like manner, for the matching relationship of other types, different parameters also can correspondingly be set set up corresponding impulsive noise model.
5. the reconstruction step of the filtering of noise data and missing data has wherein used 5 linear smoothing algorithms that each subsequence is carried out noise reduction process, and smoothing algorithm can be represented by formula (2):
Wherein x (n) is pending source sequence, and w (m) is the weight coefficient of x (m), and
L is level and smooth scope, gets L=2 herein, and y (n) is the sequence after handling.
Fig. 9 has shown the center point P (P of a certain semantic chunk
x, P
y, P
z) state before the processing of the sequence that generated after state (noise sequence that contains peak value) and the processing, experiment shows that 5 linear smoothing algorithms can be realized the noise reduction process of impulsive noise model effectively.
6. the sequence after the noise reduction process oppositely is reconstructed into position and the attitude of semantic chunk in three dimensions, the filtering of the noise data of realization movement capturing data and the reconstruction processing of missing data.Figure 10 shows that the experimental result that a certain group of exercise data that moves handled, left part is the signature point data before handling, and right part is the result after handling.Figure 11 has shown the result of partial frame in the exercise data, from a left side and right be respectively jump over, rope skipping, dancing 1, dancing 2 and the action walked.When the raw data of gathering satisfies following condition:
(1) the same signature point consecutive miss of the exercise data of Cai Jiing frame number is less than 10;
(2) on the same rigid structure body frame number of two above signature points of consecutive miss less than 10; The data processing method that proposes among the present invention can realize the accurate processing of all frame data, and handling rate reaches 100%.
Below in conjunction with the example in the accompanying drawing, be data source with the dance movement exercise data of one group of 32 signature point, the specific embodiment of the present invention is further elaborated.
At first, logical optics capture system devices collect data is 300 frames, about 5 seconds of time remaining, sample frequency 60fps.Be 3.0GHz by method proposed by the invention a CPU frequency then, realize on the PC of internal memory 1G.Operating system is WindowsXP.Based on C Plus Plus and OpenGL shape library exploitation optics human body motion capture data handling system, case verification data processed result.Concrete implementation step is:
Step 1: gather raw data and masterplate data.The original data storage form is as follows:
FRAME1
MARKER0 -268.434253-140.386279?1649.468816
MARKER1 -265.47517 200.876734?1408.155767
MARKER2 -373.405425-232.048548?1400.100957
MARKER3 -216.285255-32.192661 1366.558122
……
FRAME2
……
……
Wherein FRAMEn is expressed as the n frame data
MARKERm represents m signature point, and three numbers are subsequently represented its x respectively, y, z coordinate, the mm of unit.
Signature point sequence number in the manual demarcation masterplate data.
Step 2: be written into the masterplate data, and calculate generation standard semantic piece and standard semantic model.All 32 signature points are divided into 11 standard semantic pieces, and as shown in Figure 2, wherein, semantic build, chest, left hand, the right hand, waist, left foot and right crus of diaphragm are the rigidity quadrilateral structure, and left arm, right arm, left leg and four semantic chunks of right leg are the line segment type structure.
Step 3: be written into exercise data.
Step 4: all frame data that circulate, all semantic chunks that circulate are used the rigid structure matching algorithm, generate all alternative semantic chunks of each semantic chunk, and are stored as alternative semantic chunk sequence.This process can be described as with pseudo-code:
FOR?i＝1?to?nFrame
{
FOR?j＝1?to?11
{
A＝RIGIDMATCHING(j)
ADD?A?TO?SEQUENCE[i][j]
}
}
Step 5: according to human body physical structural characteristic and semantic model, three types semantic constraint condition described in the structure summary of the invention.And each element in the alternative semantic chunk sequence of application constraint conditional test, obtain the most rational one group of semantic chunk.
For example: bigger variation can not take place in the distance between piece " chest " and piece " waist " central point in motion process, and definition error allowed band is judged the distance between these two parts, compares with the standard semantic model, whether satisfies constraint condition.
Piece " chest " is in normal direction forward one side of piece " waist " simultaneously, and structure grain constraint condition judges whether the element in the alternative semantic chunk satisfies condition.In like manner, construct other semantic constraint conditions.
Comprehensive each semantic constraint condition finally obtains one group and has the most alternative semantic chunk of reasonable mutual relationship, with it as next step the impulsive noise model based that will set up.
Step 6:,, set up the time series of parameter respectively at the position and the attitude of each semantic chunk according to the described method of summary of the invention the 4th joint.Rigidity quadrilateral structure wherein is as " head ", " waist " etc., with center point P (P
x, P
y, P
z), normal vector N (N
x, N
y, N
z) and point to vectorial D (D
x, D
y, D
z) definition, set up 9 time serieses respectively, the line segment type structure is as " left arm ", " left leg " etc., with center point P (P
x, P
y, P
z) and point to vectorial D (D
x, D
y, D
z), set up 6 time serieses respectively.
Step 7: described 5 the linear smoothing algorithms of application invention content the 5th joint, carry out noise reduction process with the subsequence of each semantic chunk.Fig. 9 has shown the center point P (P of " head " semantic chunk
x, P
y, P
z) state before the processing of the sequence that generated after state (noise sequence that contains peak value) and the processing.
Step 8: according in each sequence after the smoothing processing corresponding to the value of each time point, oppositely be reconstructed into the position and the attitude of semantic chunk, obtain the exercise data after noise data filtering and missing data reconstruction are handled, the experimental result of handling for exercise data as shown in figure 10, left part is the signature point data before handling, and right part is the result after handling.
Step 9: the exercise data output after the processing, with document form output, data layout is:
FRAME1
MARKER0 -195.19 16.16 1656.13
MARKER1 -292.02 65.14 1614.35
MARKER2 -357.60 -105.53 1601.88
MARKER3 -269.40 -140.62 1646.24
……
FRAME2
……
……
At this moment, each data cell meaning is identical with the acquired original data, and difference is fixed as 32 for signature point number in each frame, and the gauge point order is consistent with mark dot sequency in the masterplate.
The same signature point consecutive miss of the exercise data of gathering in example frame number is less than 10, and the frame number of two above signature points of consecutive miss is less than 10 on the same rigid structure body, data processing method realizes the accurate processing of all frame data, and handling rate reaches 100%.
Claims (2)
1. dynamic space-time coupling denoise processing method that is used for the human body motion capture data, comprise will based on the optics human body motion capture data handling system of C++ and openGL shape library exploitation pack into computing machine and by human body optical motion capture system equipment obtain raw data, to the step of the processing and the human motion reconstruct of raw data, it is characterized in that further comprising the steps of:
(1) noise data and the missing data that exists in the exercise data at acquired original set up the dynamic pulse noise model based on alternative semantic chunk;
(2) generate alternative semantic chunk based on the rigid structure matching algorithm, and be defined as five kinds of basic match-types, be i.e. line segment type structure, triangular structure, free quadrilateral structure, diagonal angle quadrilateral structure and rigidity quadrilateral structure;
(3) definition semantic chunk and alternative semantic chunk, and all signature points are divided into 11 semantic chunks according to the organization of human body feature, comprise: head, chest, waist, left arm, right arm, left hand, the right hand, left leg, right leg, left foot and right crus of diaphragm, wherein head, chest, waist, left hand, the right hand, left foot and right crus of diaphragm are the rigidity quadrilateral structure; Left arm, right arm, left leg and right leg are the line segment type structure;
(4) definition semantic node and each property parameters thereof, and human body topological structure relation is described based on the human body semantic model that 22 semantic nodes are set up in this definition, the property parameters of semantic node comprises: sequence number, title, degree, level, father node, length, direction and position, and wherein sequence number, title, degree, level and father node are defined as the subordinate relation attribute; Length, direction and location definition are the locus attribute;
(5) according to the semantic constraint condition of three types of human body semantic model structures, i.e. distance restraint, angle restriction and direction constrain, whether the alternative semantic chunk that generates in the determination step (2) meets constraint condition; Return the "Yes" or "No" two states respectively according to result of determination; For missing data, assignment is error identification value e, described error identification value e be one can be by the obvious value of identification of system, thereby missing data is converted into noise data, for providing prerequisite for noise reduction process with the subsequent treatment unification;
(6) generate the most rational organization of human body, promptly the most rational one group of semantic chunk;
(7) foundation is all explained each semantic chunk based on the impulsive noise model of the dynamic time sequence of semantic chunk position and attitude attribute with position in the three dimensions and attitude, and wherein the position of rigidity quadrilateral structure can be by center point P (P
x, P
y, P
z) define, attitude can be by a normal vector N (N
x, N
y, N
z) and the vectorial D (D of sensing
x, D
y, D
z) determine; The line segment type structure can be by center point P (P
x, P
y, P
z) and the vectorial D (D of sensing
x, D
y, D
z) define, each semantic chunk is represented by 6 or 9 parameters; In the time domain scope, the value of corresponding parameters constitutes a time series in all frames, and each semantic chunk can be represented by 6 or 9 sub-time serieses; All use sub-time sequence to represent all semantic chunks, just constructed impulsive noise model based on the dynamic time sequence of semantic chunk position and attitude attribute;
(8) 5 linear smoothing algorithms are applied on the time series that step (7) set up, realize the filtering of noise data and the reconstruction processing of missing data.
2. the dynamic space-time coupling denoise processing method that is used for the human body motion capture data according to claim 1, it is characterized in that the reconstruction processing described in the step (8), be that the value of subsequence on each time point with each semantic chunk after handling oppositely is reconstructed into position and the attitude of semantic chunk in three dimensions, thus the exercise data after obtaining handling.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100101702A CN101216952B (en) | 2008-01-17 | 2008-01-17 | Dynamic spatiotemporal coupling denoise processing method for data catching of body motion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100101702A CN101216952B (en) | 2008-01-17 | 2008-01-17 | Dynamic spatiotemporal coupling denoise processing method for data catching of body motion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101216952A CN101216952A (en) | 2008-07-09 |
| CN101216952B true CN101216952B (en) | 2011-05-18 |
Family
ID=39623380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008100101702A Active CN101216952B (en) | 2008-01-17 | 2008-01-17 | Dynamic spatiotemporal coupling denoise processing method for data catching of body motion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101216952B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101477703B (en) * | 2008-10-11 | 2011-09-14 | 大连大学 | Human body animation process directly driven by movement capturing data based on semantic model |
| CN101840508B (en) * | 2010-04-26 | 2013-01-09 | 中国科学院计算技术研究所 | Method and system for automatically identifying characteristic points in human body chain structure. |
| CN102486816A (en) * | 2010-12-02 | 2012-06-06 | 三星电子株式会社 | Device and method for calculating human body shape parameters |
| CN102289801B (en) * | 2011-05-16 | 2013-08-21 | 大连大学 | Data repairing method and system for motion capture and motion capture system |
| TWI443587B (en) * | 2011-05-30 | 2014-07-01 | Univ Nat Cheng Kung | Three dimensional dual-mode scanning apparatus and three dimensional dual-mode scanning system |
| CN104103075A (en) * | 2014-07-24 | 2014-10-15 | 北京邮电大学 | Multi-view human body part semantic matching method and device |
| CN106447618B (en) * | 2016-05-20 | 2019-04-12 | 北京九艺同兴科技有限公司 | A kind of human action sequence noise-reduction method dictionary-based learning |
| CN106203480B (en) * | 2016-06-29 | 2019-06-04 | 华侨大学 | Nonlinear feature extraction and classification method based on data incomplete |
| CN107330967B (en) * | 2017-05-12 | 2020-07-24 | 武汉商学院 | Rider motion posture capturing and three-dimensional reconstruction system based on inertial sensing technology |
| CN112163516B (en) * | 2020-09-27 | 2024-05-10 | 深圳市悦动天下科技有限公司 | Rope skipping counting method and device and computer storage medium |
| CN112579565B (en) * | 2020-11-30 | 2023-04-18 | 贵州力创科技发展有限公司 | Data model management method and system of data analysis engine |
| CN113674224A (en) * | 2021-07-29 | 2021-11-19 | 浙江大华技术股份有限公司 | Monitoring point position management method and device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1924932A (en) * | 2006-09-08 | 2007-03-07 | 中国科学院计算技术研究所 | Method for correcting noises and errors in human sports trapped data |
-
2008
- 2008-01-17 CN CN2008100101702A patent/CN101216952B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1924932A (en) * | 2006-09-08 | 2007-03-07 | 中国科学院计算技术研究所 | Method for correcting noises and errors in human sports trapped data |
Non-Patent Citations (2)
| Title |
|---|
| 黄波士，张福民，张金剑.一种改进算法的光学运动捕捉***.同济大学学报（自然科学版）33 10.2005,33(10),全文. |
| 黄波士,张福民,张金剑.一种改进算法的光学运动捕捉***.同济大学学报(自然科学版)33 10.2005,33(10),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101216952A (en) | 2008-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101216952B (en) | Dynamic spatiotemporal coupling denoise processing method for data catching of body motion | |
| CN110222665B (en) | Human body action recognition method in monitoring based on deep learning and attitude estimation | |
| Mori et al. | Recovering 3d human body configurations using shape contexts | |
| CN102184541B (en) | Multi-objective optimized human body motion tracking method | |
| JP2023502795A (en) | A real-time system for generating 4D spatio-temporal models of real-world environments | |
| Ran et al. | Applications of a simple characterization of human gait in surveillance | |
| CN103003846B (en) | Articulation region display device, joint area detecting device, joint area degree of membership calculation element, pass nodular region affiliation degree calculation element and joint area display packing | |
| CN102622766A (en) | Multi-objective optimization multi-lens human motion tracking method | |
| WO2020102042A1 (en) | Physical activity quantification and monitoring | |
| CN103227888B (en) | A kind of based on empirical mode decomposition with the video stabilization method of multiple interpretational criteria | |
| CN102800126A (en) | Method for recovering real-time three-dimensional body posture based on multimodal fusion | |
| Zhu et al. | Motion capture data denoising based on LSTNet autoencoder | |
| CN102682452A (en) | Human movement tracking method based on combination of production and discriminant | |
| CN106815855A (en) | Based on the human body motion tracking method that production and discriminate combine | |
| Uddin et al. | Human Activity Recognition via 3-D joint angle features and Hidden Markov models | |
| Fan et al. | Revitalizing optimization for 3d human pose and shape estimation: A sparse constrained formulation | |
| CN101533526B (en) | Optical motion capture data processing method based on dynamic template | |
| CN116700471A (en) | Method and system for enhancing user experience of virtual reality system | |
| Schreiner et al. | Global position prediction for interactive motion capture | |
| Dhore et al. | Human Pose Estimation And Classification: A Review | |
| CN109993818B (en) | Method, device, equipment and medium for synthesizing motion of three-dimensional human body model | |
| CN115222882A (en) | Model construction method and device based on human motion video and storage medium | |
| Ugolotti et al. | Differential evolution based human body pose estimation from point clouds | |
| Zhu | Refining method of mocap data based on LSTM | |
| Chen et al. | An integrated sensor network method for safety management of construction workers |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |