CN104050461A - Complex 3D motion recognition method and device - Google Patents

Abstract

The invention provides a complex 3D motion recognition method and device. Multiple track points are obtained by sampling a motion track firstly, then category and dimension information of each track point is obtained, the category and dimension information of all the track points is sequentially arranged to serve as a descriptor of the whole motion track, consequently, feature information of the motion track is obtained, matching with a database is carried out according to the feature information, motion types after matching is conducted are obtained, and motion recognition is completed. According to the method and device, distinguishing features of the motion track can be effectively used for motion recognition, and therefore 3D complex motion recognition precision and accuracy are improved.

Description

A kind of complicated 3D motion recognition methods and device

Technical field

The present invention relates to shape motion discriminance analysis field, relate in particular to a kind of complicated 3D motion recognition methods and device.

Background technology

Motion discriminance analysis based on track is popular research topic in robot and automatic field always.This research topic is developed so far, and complicated motion is identified and remained the huge challenge that people have to face in various applications.

Nowadays, huge effect is just being brought into play in more and more application based on motion identification in engineering field, such as video monitoring, automatically monitoring and identification, man-machine interaction etc.

For good motion recognition effect, good movement locus model and efficient identification engine are indispensable.Because movement locus is the important clue of carrying out recognizing model of movement, and uses good locus model more to contribute to extract outstanding motion feature.Meanwhile, in order to take into account speed and the precision of identification, must there is efficient identification engine to coordinate the work of locus model.Multiple means and the instruments such as Euclidean distance, the description of track profile, various transforming function transformation functions have been used in existing a lot of motion recognition methodss.But these methods all can not effectively be utilized the notable feature of movement locus substantially, therefore, in most cases can not be used for identifying complicated movement locus.

Therefore,, for above-mentioned technical matters, be necessary to provide a kind of and novel complicated 3D carried out to motion recognition methods.

Summary of the invention

The invention provides a kind of complicated 3D motion recognition methods and device, the present invention can effectively utilize the identification of moving of the notable feature of movement locus, thereby has improved 3D compound movement accuracy of identification and accuracy rate.

A kind of complicated 3D motion recognition methods, comprising:

Obtain multiple tracing points of the 3D movement locus of object;

Calculate curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point, signature descriptor is used for representing tracing point characteristic information;

According to four parameters in the signature descriptor of each tracing point and zero magnitude relationship, multiple tracing points are divided into four classes, wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

By preset algorithm, each tracing point is carried out to computing, obtain the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

The sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus, descriptor is used for representing 3D movement locus characteristic information;

The descriptor of 3D movement locus is mated with multiple descriptors in database, after the match is successful, the type of sports corresponding with the descriptor of database given to this 3D movement locus.

Preferably, before the descriptor that obtains 3D movement locus, also comprise:

By the tracing point composition track atom that in multiple tracing points, position is adjacent and classification is identical, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

Yardstick information using the yardstick information of a tracing point in each track atom as this track atom; Descriptor using the classification of each track atom and yardstick information as this track atom; The sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

Preferably, the yardstick information using the yardstick information of a tracing point in each track atom as this track atom comprises:

Investigate the yardstick information of each tracing point in each track atom, the yardstick information using maximum yardstick information as this track atom.

Preferably, by preset algorithm, each tracing point is carried out to computing, the yardstick information that obtains each tracing point comprises:

For each tracing point in Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind;

For each tracing point in the 3rd class and the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class;

For each tracing point in the first kind, calculate by one of above-mentioned two kinds of modes, obtain the yardstick information of each tracing point in the first kind.

Preferably, the descriptor of this 3D movement locus is mated and is comprised with the descriptor of database:

The descriptor of the descriptor of 3D movement locus and database is carried out to LCSS calculation of parameter one by one, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database;

Obtain the corresponding descriptor of minimum value in multiple distances, using this descriptor as matching result.

A kind of complicated 3D motion recognition device, comprising:

Acquiring unit, for obtaining multiple tracing points of 3D movement locus of object;

Signature descriptor unit, for calculating curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point;

Taxon, for four parameters of signature descriptor according to each tracing point and zero magnitude relationship, is divided into four classes by multiple tracing points, and wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

Descriptor unit, for each tracing point being carried out to computing by preset algorithm, obtains the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

The one 3D descriptor unit, for the sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus;

Matching unit for multiple to the descriptor of 3D movement locus and database descriptors are mated, is given the type of sports corresponding with the descriptor of database this 3D movement locus after the match is successful.

Preferably, also comprise:

Track atomic unit, for tracing point adjacent multiple tracing points position and that classification is identical is formed to track atom, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

The 2nd 3D descriptor unit, for the yardstick information using the yardstick information of a tracing point of each track atom as this track atom; Descriptor using the classification of each track atom and yardstick information as this track atom; The sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

Preferably, the 2nd 3D descriptor unit, also for investigating the yardstick information of the each tracing point of each track atom, the yardstick information using maximum yardstick information as this track atom.

Preferably, descriptor unit comprises:

The first computing unit, for for the each tracing point of Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind;

The second computing unit, for for the 3rd class and the each tracing point of the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class;

The first computing unit or the second computing unit, also for calculating for the each tracing point of the first kind, obtain the yardstick information of each tracing point in the first kind.

Preferably, matching unit also comprises:

LCSS computing unit, carries out LCSS calculation of parameter one by one by the descriptor of the descriptor of 3D movement locus and database, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database;

Contrast unit, for obtaining the corresponding descriptor of multiple distance minimum value, using this descriptor as matching result.

The invention provides a kind of complicated 3D motion recognition methods and device; the present invention's movement locus of first sampling obtains multiple tracing points; then obtaining respectively classification and the yardstick information of each tracing point; each tracing point classification and yardstick information are arranged in order to the descriptor as whole movement locus; thereby obtain the characteristic information of movement locus; mate with database according to characteristic information, thereby obtain the type of sports after coupling, complete motion identification.The present invention can effectively utilize the identification of moving of the notable feature of movement locus, thereby has improved 3D compound movement accuracy of identification and accuracy rate.

Brief description of the drawings

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only embodiments of the invention, for those of ordinary skill in the art, do not paying under the prerequisite of creative work, other accompanying drawing can also be provided according to the accompanying drawing providing.

Fig. 1 is the complicated 3D motion of the disclosed one of embodiment of the present invention recognition methods process flow diagram;

Fig. 2 is disclosed another the complicated 3D motion recognition methods process flow diagram of the embodiment of the present invention;

Fig. 3 is the structural representation of the disclosed a kind of complicated 3D motion recognition device of the embodiment of the present invention;

Fig. 4 is the structural representation of disclosed another the complicated 3D motion recognition device of the embodiment of the present invention;

Fig. 5 is the structural representation of disclosed another the complicated 3D motion recognition device of the embodiment of the present invention;

Fig. 6 is the structural representation of disclosed another the complicated 3D motion recognition device of the embodiment of the present invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiment.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtaining under creative work prerequisite, belong to the scope of protection of the invention.

As shown in Figure 1, the invention provides a kind of complicated 3D motion recognition methods, comprising:

Step S101: the multiple tracing points that obtain the 3D movement locus of object;

The tracing point that obtains movement locus of object can have two kinds of modes, the first is: the position that gathers objects in images, using the position of object in every two field picture as a tracing point, gather successively multiple image and can obtain multiple tracing points, multiple image forms the movement locus of object, is the 3D movement locus of object.The second is: first obtain the 3D movement locus of object, then according to certain frequency, 3D movement locus is sampled, the tracing point that multiple tracing points of continuous acquisition are used in the present invention.

The quantity of tracing point is at least one, is as the criterion with the movement locus of complete representation object.

Step S102: calculate curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point; Signature descriptor is used for representing the characteristic information of tracing point.

To the multiple tracing points that obtain in above-mentioned steps, calculate respectively curvature and the moment of torsion of each tracing point, and calculate the first order derivative of curvature and the first order derivative of moment of torsion, and be referred to as curvature derivative and moment of torsion derivative, curvature shows the numerical value of curve at certain any degree of crook on mathematics, curvature is larger, represents that the degree of crook of curve is larger.Moment of torsion shows the numerical value of cambered surface at certain any degree of crook on mathematics, and moment of torsion is larger, represents that the degree of crook of cambered surface is larger.

Because the type of sports of object can be divided into straight line, curve, left hand helix cambered surface and four kinds of situations of right-handed helix cambered surface, each tracing point is above-mentioned type of sports one, therefore distinguish the sports category of each tracing point, just can draw the type of sports of whole track, so need the clear sports category of distinguishing each tracing point, the present invention adopts curvature, moment of torsion, curvature derivative and moment of torsion derivative to distinguish.

Step S103: according to four parameters in the signature descriptor of each tracing point and zero magnitude relationship, multiple tracing points are divided into four classes, wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

As shown in table 1, be the relation table between classification and four parameters, wherein, k represents that curvature, τ represent moment of torsion, k _srepresent curvature derivative, τ _srepresent moment of torsion derivative, the first kind represents that for representing straight line, Equations of The Second Kind curve, the 3rd class represent that left hand helix cambered surface, the 4th class represent right-handed helix cambered surface, and "/" represents not exist this content.

Table 1

Class	k	τ	k s	τ s
					The first kind	0	0	0	0
Equations of The Second Kind	≠0	0	/	0
					The 3rd class	≠0	<0	/	/
The 4th class	≠0	>0	/	/

Wherein, the first kind is straight line, and straight curvature of a curve and moment of torsion are all zero, so curvature derivative and moment of torsion derivative are all zero; Equations of The Second Kind is curve, and bent curvature of a curve is non-vanishing, but is zero to the curvature derivative obtaining after curvature differentiate, and curve does not have moment of torsion, so moment of torsion and moment of torsion derivative are all nothing; The 3rd class is left hand helix cambered surface, and the curvature of cambered surface is non-vanishing constant, and the curvature derivative after differentiate is zero, and it is right-handed helix cambered surface that the moment of torsion of left hand helix cambered surface is less than zero, the four class, consistent with the content of left hand helix cambered surface, and just moment of torsion is greater than zero.

Have foregoing known, four concrete numerical value of parameter of each classification are inconsistent, so by judging four parameters of each tracing point and zero magnitude relationship, can accurately obtain the classification of tracing point.Concrete, when being all zero, four parameters judge that tracing point is as the first kind, non-vanishing when curvature, moment of torsion is zero, judges that tracing point is as Equations of The Second Kind, judges that when moment of torsion is less than zero tracing point is as the 3rd class, when moment of torsion is greater than zero, judges that tracing point is as the 4th class.Can accurately draw the classification information of each tracing point according to above-mentioned decision rule.

Can not draw accurately separately the movement locus of object according to classification information, such as being all the track of straight line, curve, left hand helix cambered surface and right-handed helix cambered surface, but in the time that the degree of crook of curve or curved surface is inconsistent, can not accurately draw the movement locus of object, therefore also need to carry out further basis for estimation according to yardstick information, yardstick information is a parameter that represents object degree of crook.

Step S104: by preset algorithm, each tracing point is carried out to computing, obtain the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

Each tracing point is calculated, obtain the yardstick information of each tracing point, concrete:

For each tracing point in Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind; Because Equations of The Second Kind is curve, curve only has curvature there is no moment of torsion, so adopt curvature and curvature derivative to calculate yardstick information, curvature and curvature derivative are all the parameters that represents curve degree of crook, so yardstick information can represent the degree of crook of this tracing point.

For each tracing point in the 3rd class and the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class; In the 3rd class and the 4th class, to moment of torsion and moment of torsion derivative calculations yardstick information, moment of torsion can accurately represent the degree of crook of cambered surface, so the degree of crook that the yardstick information obtaining can accurate response cambered surface.

For each tracing point in the first kind, calculate by one of above-mentioned two kinds of modes, obtain the yardstick information of each tracing point in the first kind.Because four parameters corresponding to the first kind are all zero, so both can adopt the account form of Equations of The Second Kind, also can adopt the account form of the 3rd class and the 4th class, be all zero no matter adopt the yardstick information that any account form draws.

Obtain after the classification information and yardstick information of each tracing point, classification information and yardstick information are carried out to combination, as the signature descriptor of this tracing point, the i.e. characteristic information of this tracing point.

Step S105: the sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus; Descriptor is used for representing 3D motion characteristics information.

The combination of the signature descriptor of multiple tracing points is the signature descriptor of whole 3D movement locus, i.e. the characteristic information of whole movement locus, so far just by the 3D movement locus of object, is converted to the accurately characteristic information of identification of computing machine completely.

Step S106: the descriptor of 3D movement locus is mated with multiple descriptors in database, the type of sports corresponding with the descriptor of database given to this 3D movement locus after the match is successful.

Pre-stored in database have multiple descriptors, also has the type of sports corresponding with multiple descriptors, for example: keep straight on, turn left curved, turn right curved, go upstairs etc.

The descriptor of the descriptor of 3D movement locus and database is carried out to LCSS calculation of parameter one by one, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database; Obtain the corresponding descriptor of minimum value in multiple distances, using this descriptor as matching result.Thereby complete motion identification.

The invention provides a kind of complicated 3D motion recognition methods and device; the present invention's movement locus of first sampling obtains multiple tracing points; then obtaining respectively classification and the yardstick information of each tracing point; each tracing point classification and yardstick information are arranged in order to the descriptor as whole movement locus; thereby obtain the characteristic information of movement locus; mate with database according to characteristic information, thereby obtain the type of sports after coupling, complete motion identification.The present invention can effectively utilize the identification of moving of the notable feature of movement locus, thereby has improved 3D compound movement accuracy of identification and accuracy rate.

In the above-described embodiments using the coupling foundation as 3D motion of each tracing point, because the process of object of which movement is a continuous process, so consistent when the sports category of multiple movement locus points and motion path, if the words that the movement locus of each tracing point is calculated, increase undoubtedly the workload of computing machine, so tracing point consistent sports category can be merged, omit unnecessary redundant information, to reduce the workload of computing machine.

As shown in Figure 2, the present invention also provides a kind of complicated 3D motion recognition methods, and on the basis of the embodiment shown in Fig. 1, delete step S105 increases following step:

Step S201: by the tracing point composition track atom that in multiple tracing points, position is adjacent and classification is identical, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

Because the sports category of multiple tracing points is identical, so by a track atom of tracing point combination identical multiple sports category, the classification of the classification of the track atom after the combination tracing point inner with it is identical, has so just reduced a large amount of redundant informations, and the workload of computing machine is die-offed.

Step S202: the yardstick information using the yardstick information of a tracing point in each track atom as this track atom;

In above-mentioned steps, obtain the classification of this track atom, in this step, need to obtain the yardstick information of track atom, can choose at random the yardstick information of any tracing point in track atom as the yardstick information of this track atom, but the yardstick information of the each tracing point in track atom varies, preferably, can investigate the yardstick information of each tracing point in each track atom, the yardstick information using maximum yardstick information as this track atom.

Step S203: the descriptor using the classification of each track atom and yardstick information as this track atom;

Step S204: the sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

The descriptor that is formed track atom by the classification of track atom and yardstick information, carries out assembled arrangement successively by the descriptor of multiple track atoms, just can obtain the descriptor of whole 3D movement locus.The descriptor of this 3D movement locus, for the descriptor of the 3D movement locus in step S105, because redundant information wherein reduces, so improved efficiency in matching process.

All the other steps in the present embodiment are consistent with the step shown in Fig. 1, do not repeat them here.

A kind of specific embodiment of complicated 3D motion recognition methods is provided below:

1, calculate curvature k and the moment of torsion τ of each point on track;

2, ask for each tracing point curvature and the torque parameter first order derivative k for arc length _sand τ _s, by the first order derivative of the first order derivative of curvature, moment of torsion, curvature and moment of torsion, totally four parameters form the signature descriptor of each tracing point.

The expression mode of a 3D movement locus is Γ (t)={ X (t), Y (t), Z (t) | t ∈ [1, N] }, wherein Γ represents movement locus, t is the time of every two field picture, and N represents that the length of track is the time of last pin image, and wherein N is natural number; In order to represent complicated track efficiently, introduce the original expression mode that signature descriptor replaces tracing point:

S＝{k(t),k _s(t),τ(t),τ _s(t)|t∈[1,N]}

Four parameters in signature descriptor are respectively: curvature k, and moment of torsion τ, and they are about the first order derivative k of arc length s _sand τ _s.Parameter in signature descriptor can be used for offering help for 3D motion identification, but because its redundant information is too much, can not meets directly overall track is carried out to the efficient requirement of describing.

3, investigate successively the value of the parameter of composition each point signature descriptor;

According to curvature k, moment of torsion τ and first order derivative k thereof _sand τ _smagnitude relationship with 0, is divided into A, B, C, D tetra-classes according to certain rule by tracing point.

In above-mentioned steps, we successfully use signature descriptor S to represent each tracing point, next just need to each tracing point be classified according to the parameter of signature descriptor, and classifying rules is as follows:

Class	k	τ	k s	τ s
					A	0	0	0	0
B	≠0	0	/	0
					C	≠0	<0	/	/
D	≠0	>0	/	/

Wherein four kinds represent respectively: category-A, straight line; Category-B, plane arc; C class, left hand helix; D class, right-handed helix.

4, the classification of the continuous tracing point of inspecting position, forms a track atom by similar continuous path point, and movement locus can be made up of some track atoms; Using the classification of the tracing point of composition track atom as the former subclass of this track.

In the time that a series of continuous tracing points have identical classification, they must be of similar shape feature, and we just represent the orbit segment at these places with a track atom.Meanwhile, the classification of these points is assigned to corresponding track atom by we.Such method of operating just makes whole track to be represented by several different classes of track atoms.This method for expressing represents that with putting the method for whole movement locus is more simple compared to original, assesses the cost less.

If the track atom that movement locus has m section represents that mode is as follows:

Γ＝{Prim(1),Prim(2),...,Prim(i),...,Prim(m)}

Prim(i)＝{p(n _i),p(n _i+1),...,p(n _i+j),...,p(n _i+l _i-1)}

Entirety track Γ is represented by the sequence of track atom Prim (i), and track atom is by a series of tracing points composition, n here _ifirst point of i track atom of representative composition, l _irepresent the length of i track atom.

5, the tracing point of all category-Bs is asked for to its yardstick SP (B) according to certain rule, and form the S-IGS descriptor of this point together with its classified information;

The tracing point of all C classes or D class is asked for to its yardstick SP (C), SP (D) according to certain rule, and form the S-IGS descriptor of this point together with its classified information.

Represent track by the sorting technique of step S3 merely, can cause dissimilar movement locus to there is identical title, cause the decline of matching precision.In order to make up this shortcoming, we introduce supplementary-yardstick SP, in order to differentiate the track atom with same kind.

We are each tracing point divided rank according to the signature descriptor parameter of each tracing point, as its yardstick.Wherein, also difference to some extent of the grade classification mode of the point of four kinds.Because category-A is straight-line segment, the yardstick of each point is identical, therefore without asking for.The acquiring method of B, C, D class is as follows:

$\mathrm{SP}\left(B\right)=\sqrt{k^2+\mathrm{\λ}*k_s^2}$

$\mathrm{SP}(C/D)=\sqrt{{\mathrm{\τ}}^2+\mathrm{\λ}*{\mathrm{\τ}}_s^2}$

The λ be here one we can according to application need regulate parameter, it is directly proportional to the complexity of track conventionally, generally we get λ=1.

6, investigate the yardstick information of the tracing point of composition track atom, get the maximal value wherein yardstick as this track atom; The IGS descriptor of track atom is combined with yardstick information S, the S-IGS descriptor of composition track atom;

The dimension calculation method of track atom Prim (i) is as follows:

${\mathrm{SP}}_B\left(i\right)=\max \{{\sqrt{k_{p(n_i+j)}^2+\mathrm{\λ}*k_{s\·p(n_i+j)}^2}\}}_{j=0}^{l_i-1}$

${\mathrm{SP}}_{C/D}\left(i\right)=\max \{{\sqrt{{\mathrm{\τ}}_{p(n_i+j)}^2+\mathrm{\λ}*{\mathrm{\τ}}_{s\·p(n_i+j)}^2}\}}_{j=0}^{l_i-1}$

At this moment, track atom has just had the dual label of classified information (IGS descriptor) and yardstick (S).For example, the track atom of a category-B has the yardstick of SP3 simultaneously, and we just represent this track atom with label " B3 ".Like this, of a sort track atom can have different numerical information in label, has strengthened resolution characteristic and matching precision.

7, movement locus is divided into some sections naturally by track atom, track atom of every section of correspondence; The S-IGS descriptor of each track atom is arranged in to a sequence in order, with this sequence S-IGS descriptor of movement locus as a whole;

According to above step, a complete movement locus can be showed by the track atomic series with S-IGS descriptor by several, has wherein both comprised classified information label A BCD, also comprised yardstick numeral 1,2 ... K, K is natural number.

The S-IGS descriptor of the each track atom in complete trajectory is formed a sequence by we, as the S-IGS descriptor of this track, that is the title of this track.For belonging to dissimilar movement locus, their corresponding track atoms have different yardsticks.Like this, we just can distinguish the track with same category information labels by the yardstick information in title.

8, the track descriptor in target trajectory S-IGS descriptor and coupling storehouse is carried out to LCSS calculation of parameter one by one; Get and a track of target trajectory LCSS calculated value minimum, think that target trajectory and this track belong to same movement locus.

LCSS algorithm can be looked for similar part efficiently between the title (S-IGS descriptor) of two tracks.The distance function D of two track Q (n) that contain respectively the track P (m) of m track atom and contain n track atom based on LCSS algorithm _{δ, ε}{ P (m), Q (n) } is defined as follows:

$D_{\mathrm{\&delta;},\mathrm{\&epsiv;}}\{P\left(m\right),Q\left(n\right)\}=\left\{\begin{array}{c}0,\mathrm{iflengthofPorQis}0\\ D_{\mathrm{\&delta;},\mathrm{\&epsiv;}}\{P(m-1),Q(n-1)\}+1,\\ \mathrm{ifd}\{{\mathrm{Prim}}_{p\left(m\right)},\mathrm{Pri}{m}_{Q\left(n\right)}\}<\mathrm{\&epsiv;and}|m-n|<\mathrm{\&delta;}\\ \max \{D_{\mathrm{\&delta;},\mathrm{\&epsiv;}}P\left(m\right),Q(n-1),D_{\mathrm{\&delta;},\mathrm{\&epsiv;}}P(m-1),Q\left(n\right)\},\\ \mathrm{otherwise}\end{array}\right.$

The parameter ε is here the similarity distance threshold value of two track atoms.If two interatomic distances of track are less than ε, can think that the orbit segment at these two track atom places is similar parts.δ is the parameter for controlling the positional distance of the track atom that can mutually mate in track atomic series.P (m-1) represents that track p (m) removes the track of last track atom place section.D{Prim _{p (m)}, Prim _{q (n)}be defined as follows:

$d\{\mathrm{Pri}{m}_{p\left(m\right)},\mathrm{Pri}{m}_{Q\left(n\right)}\}=\sqrt{(\mathrm{\&Delta;}{k}_{P,Q}^2+\mathrm{\&lambda;}*\mathrm{\&Delta;}{k}_{s\&CenterDot;P,Q}^2)*(\mathrm{\&Delta;}{\mathrm{\&tau;}}_{P,Q}^2+\mathrm{\&lambda;}*\mathrm{\&Delta;}{\mathrm{\&tau;}}_{s\&CenterDot;P,Q}^2)}$

Wherein,

$\left\{\begin{array}{c}\mathrm{\&Delta;}{k}_{P,Q}^2={(k_{P\left(m\right)}-k_{Q\left(n\right)})}^2\\ \mathrm{\&Delta;}{k}_{s\&CenterDot;P,Q}^2={(k_{s\&CenterDot;P\left(m\right)}-k_{s\&CenterDot;Q\left(n\right)})}^2\\ \mathrm{\&Delta;}{\mathrm{\&tau;}}_{P,Q}^2={({\mathrm{\&tau;}}_{P\left(m\right)}-{\mathrm{\&tau;}}_{Q\left(n\right)})}^2\\ \mathrm{\&Delta;}{\mathrm{\&tau;}}_{s\&CenterDot;P,Q}^2={({\mathrm{\&tau;}}_{s\&CenterDot;P\left(m\right)}-{\mathrm{\&tau;}}_{s\&CenterDot;Q\left(n\right)})}^2\end{array}\right.$

$\{k_{P\left(m\right)},k_{s\&CenterDot;P\left(m\right)},{\mathrm{\&tau;}}_{P\left(m\right)},{\mathrm{\&tau;}}_{s\&CenterDot;P\left(m\right)}\}=\{k_{P(n_m+j)},k_{s\&CenterDot;P(n_m+j)},{\mathrm{\&tau;}}_{P(n_m+j)},{\mathrm{\&tau;}}_{s\&CenterDot;P(n_m+j)}\}$

$\left\{\begin{array}{c}{j}_B=\arg \underset{j}{\max }{\left\{\sqrt{k_{P(n_m+j)}^2+\mathrm{\&lambda;}*k_{s\&CenterDot;P(n_m+j)}^2}\right\}}_{j=0}^{l_m-1}\\ j_{C/D}=\arg \underset{j}{\max }{\left\{\sqrt{{\mathrm{\&tau;}}_{P(n_m+j)}^2+\mathrm{\&lambda;}*{\mathrm{\&tau;}}_{s\&CenterDot;P(n_m+j)}^2}\right\}}_{j=0}^{l_m-1}\end{array}\right.$

By target trajectory and coupling, the movement locus one in storehouse is asked for distance once LCSS algorithm, and get its middle distance minimum one, think that target trajectory and this movement locus belong to same and move, complete motion identification mission with this.

As can be seen from the above technical solutions, complicated 3D motion recognition methods provided by the invention is in movement locus coupling and identification, can movement locus be carried out the extraction of feature and effectively be represented, can effectively remove the redundant information in tracing point, improve accuracy rate and the efficiency of identification, realized the accurate identification of complicated 3D motion.

As shown in Figure 3, the invention provides a kind of complicated 3D motion recognition device, comprising:

Acquiring unit 100, for obtaining multiple tracing points of 3D movement locus of object;

Signature descriptor unit 200, for calculating curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point;

Taxon 300, for four parameters of signature descriptor according to each tracing point and zero magnitude relationship, is divided into four classes by multiple tracing points, and wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

Descriptor unit 400, for each tracing point being carried out to computing by preset algorithm, obtains the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

The one 3D descriptor unit 500, for the sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus;

Matching unit 600 for multiple to the descriptor of 3D movement locus and database descriptors are mated, is given the type of sports corresponding with the descriptor of database this 3D movement locus after the match is successful.

Preferably, as shown in Figure 4, the present invention also provides a kind of complicated 3D motion recognition device to comprise:

Acquiring unit 100, signature descriptor unit 200, taxon 300, descriptor unit 400,

Track atomic unit 700, for tracing point adjacent multiple tracing points position and that classification is identical is formed to track atom, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

The 2nd 3D descriptor unit 800, for the yardstick information using the yardstick information of a tracing point of each track atom as this track atom; Descriptor using the classification of each track atom and yardstick information as this track atom; The sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

The 2nd 3D descriptor unit 800, also for investigating the yardstick information of the each tracing point of each track atom, the yardstick information using maximum yardstick information as this track atom.

As shown in Figure 5, the descriptor unit 400 of complicated 3D motion recognition device comprises:

The first computing unit 401, for for the each tracing point of Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind;

The second computing unit 402, for for the 3rd class and the each tracing point of the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class;

The first computing unit 401 or the second computing unit 402, also for calculating for the each tracing point of the first kind, obtain the yardstick information of each tracing point in the first kind.

As shown in Figure 6, the matching unit 600 of complicated 3D motion recognition device also comprises:

LCSS computing unit 601, carries out LCSS calculation of parameter one by one by the descriptor of the descriptor of 3D movement locus and database, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database;

Contrast unit 602, for obtaining the corresponding descriptor of multiple distance minimum value, using this descriptor as matching result.

If the function described in the present embodiment method realizes and during as production marketing independently or use, can be stored in a computing equipment read/write memory medium using the form of SFU software functional unit.Based on such understanding, the part that the embodiment of the present invention contributes to prior art or the part of this technical scheme can embody with the form of software product, this software product is stored in a storage medium, comprise that some instructions (can be personal computers in order to make a computing equipment, server, mobile computing device or the network equipment etc.) carry out all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium comprises: USB flash disk, portable hard drive, ROM (read-only memory) (ROM, Read-Only Memory), the various media that can be program code stored such as random access memory (RAM, Random Access Memory), magnetic disc or CD.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is and the difference of other embodiment, between each embodiment same or similar part mutually referring to.

To the above-mentioned explanation of the disclosed embodiments, make professional and technical personnel in the field can realize or use the present invention.To be apparent for those skilled in the art to the multiple amendment of these embodiment, General Principle as defined herein can, in the situation that not departing from the spirit or scope of the present invention, realize in other embodiments.Therefore, the present invention will can not be restricted to these embodiment shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.

Claims (10)

1. a complicated 3D motion recognition methods, is characterized in that, comprising:

Obtain multiple tracing points of the 3D movement locus of object;

Calculate curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point, signature descriptor is used for representing tracing point characteristic information;

According to four parameters in the signature descriptor of each tracing point and zero magnitude relationship, multiple tracing points are divided into four classes, wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

By preset algorithm, each tracing point is carried out to computing, obtain the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

The sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus, descriptor is used for representing 3D movement locus characteristic information;

The descriptor of 3D movement locus is mated with multiple descriptors in database, after the match is successful, the type of sports corresponding with the descriptor of database given to this 3D movement locus.

2. the method for claim 1, is characterized in that, before the descriptor that obtains 3D movement locus, also comprises:

By the tracing point composition track atom that in multiple tracing points, position is adjacent and classification is identical, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

Yardstick information using the yardstick information of a tracing point in each track atom as this track atom; Descriptor using the classification of each track atom and yardstick information as this track atom; The sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

3. method as claimed in claim 2, is characterized in that, the yardstick information using the yardstick information of a tracing point in each track atom as this track atom comprises:

Investigate the yardstick information of each tracing point in each track atom, the yardstick information using maximum yardstick information as this track atom.

4. the method for claim 1, is characterized in that, by preset algorithm, each tracing point is carried out to computing, and the yardstick information that obtains each tracing point comprises:

For each tracing point in Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind;

For each tracing point in the 3rd class and the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class;

For each tracing point in the first kind, calculate by one of above-mentioned two kinds of modes, obtain the yardstick information of each tracing point in the first kind.

5. the method for claim 1, is characterized in that, the descriptor of this 3D movement locus is mated and comprised with the descriptor of database:

The descriptor of the descriptor of 3D movement locus and database is carried out to LCSS calculation of parameter one by one, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database;

Obtain the corresponding descriptor of minimum value in multiple distances, using this descriptor as matching result.

6. a complicated 3D motion recognition device, is characterized in that, comprising:

Acquiring unit, for obtaining multiple tracing points of 3D movement locus of object;

Signature descriptor unit, for calculating curvature and the moment of torsion of each tracing point, the curvature of calculating each tracing point obtains curvature derivative for the first order derivative of arc length, and the moment of torsion that calculates each tracing point obtains moment of torsion derivative for the first order derivative of arc length; By the curvature of each tracing point, moment of torsion, curvature derivative and four parameters of moment of torsion derivative, form the signature descriptor of each tracing point;

Taxon, for four parameters of signature descriptor according to each tracing point and zero magnitude relationship, is divided into four classes by multiple tracing points, and wherein the first kind is that straight line, Equations of The Second Kind are that plane arc, the 3rd class are that left hand helix, the 4th class are right-handed helix;

Descriptor unit, for each tracing point being carried out to computing by preset algorithm, obtains the yardstick information of each tracing point, by the combination of the signature descriptor of each tracing point and yardstick information, as the descriptor of this tracing point; Described yardstick information is used to indicate the degree of crook of different tracks point;

The one 3D descriptor unit, for the sequence that the descriptor of multiple tracing points is combined successively, as the descriptor of described 3D movement locus;

Matching unit for multiple to the descriptor of 3D movement locus and database descriptors are mated, is given the type of sports corresponding with the descriptor of database this 3D movement locus after the match is successful.

7. device as claimed in claim 6, is characterized in that, also comprises:

Track atomic unit, for tracing point adjacent multiple tracing points position and that classification is identical is formed to track atom, generate multiple track atoms, the classification of each track atom is identical with the classification of tracing point in it, comprise the first kind, Equations of The Second Kind the 3rd class and the 4th class, wherein each track atom comprises at least one tracing point;

The 2nd 3D descriptor unit, for the yardstick information using the yardstick information of a tracing point of each track atom as this track atom; Descriptor using the classification of each track atom and yardstick information as this track atom; The sequence that the descriptor of multiple track atoms is combined successively, as the descriptor of described 3D movement locus.

8. device as claimed in claim 7, is characterized in that, the 2nd 3D descriptor unit, also for investigating the yardstick information of the each tracing point of each track atom, the yardstick information using maximum yardstick information as this track atom.

9. device as claimed in claim 6, is characterized in that, descriptor unit comprises:

The first computing unit, for for the each tracing point of Equations of The Second Kind, by curvature and square root sum square curvature derivative, as the yardstick information of the each tracing point in Equations of The Second Kind;

The second computing unit, for for the 3rd class and the each tracing point of the 4th class, by square root sum square of moment of torsion and moment of torsion derivative, as the yardstick information of each tracing point in the 3rd class and the 4th class;

The first computing unit or the second computing unit, also for calculating for the each tracing point of the first kind, obtain the yardstick information of each tracing point in the first kind.

10. device as claimed in claim 6, is characterized in that, matching unit also comprises:

LCSS computing unit, carries out LCSS calculation of parameter one by one by the descriptor of the descriptor of 3D movement locus and database, multiple distances of multiple descriptors in the descriptor of acquisition 3D movement locus and database;

Contrast unit, for obtaining the corresponding descriptor of multiple distance minimum value, using this descriptor as matching result.