CN106815855A - Based on the human body motion tracking method that production and discriminate combine - Google Patents
Based on the human body motion tracking method that production and discriminate combine Download PDFInfo
- Publication number
- CN106815855A CN106815855A CN201510870921.8A CN201510870921A CN106815855A CN 106815855 A CN106815855 A CN 106815855A CN 201510870921 A CN201510870921 A CN 201510870921A CN 106815855 A CN106815855 A CN 106815855A
- Authority
- CN
- China
- Prior art keywords
- human body
- artis
- human
- skeleton
- image
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of human body motion tracking method combined based on production and discriminate, mainly solve the problems, such as that mans motion simulation result is inaccurate in the prior art.Implementation step is：Set up human skeleton model；Preprocessed video image obtains detection artis；Extract the bandlet2 features of video image；It is input into the bandlet2 features extracted and predicts human body attitude using double gauss；Human skeleton model is initialized according to the artis for detecting；Build the double gauss artis for predicting and 2D the and 3D similarity functions for detecting artis；Similarity function is minimized under bone length constraint obtain lineup's body attitude；Good athletic posture of the state minimum with previous frame skeleton Euclidean distance as present frame is selected from the lineup's body attitude for obtaining.This method is high with tracking result accuracy compared with existing human body tracing method, stability advantage high, can be used for therapeutic treatment, athletic training, cartoon making and intelligent monitor system.
Description
Invention field
The invention belongs to technical field of image processing, further relate to realize in computer vision field a kind of method of mans motion simulation, realize that mans motion simulation and 3 d pose are estimated using a kind of method of multiple-objection optimization, can be used for the fields such as athletic training and cartoon making.
Background technology
The main task of mans motion simulation is that human body contour outline is detected from video image, then the artis of human body is positioned, and human motion attitude is identified on this basis, final to rebuild 3 d human motion attitude.Because current video image is the projection on 2d of human body contour outline in three-dimensional scenic, so, lost substantial amounts of depth information, and during human motion, human limb occurs often from eclipse phenomena, there is ambiguousness in video image, this makes it difficult to recover human motion attitude from unmarked monocular video.But, because the mans motion simulation based on monocular video has potential application and economic worth in various aspects such as therapeutic treatment, athletic training, cartoon making, intelligent monitor systems, so receiving the concern of many scholars.So far, the method for the mans motion simulation based on video is broadly divided into two major classes：Mans motion simulation based on study and the mans motion simulation based on model.
The first, the human body motion tracking method based on study：The method extracts accurate characteristics of image in the video image and target video image lane database of training first, then the mapping between the characteristics of image and movement capturing data in learning training vedio data storehouse, finally directly recovers 3 d pose on target video image using characteristics of human body.Such as Urtasun et al. (R.Urtasun
and T.Darrell.Local Probabilistic Regression for
Activity-Independent Human Pose Inference IEEE Conference on Computer Vision
And Pattern Recognition (CVPR), 2008) article, exactly instruct to track 3 d human motion in monocular video sequence using balance Gaussian process dynamic model, the dynamic model is obtained from the less training data learning comprising various modes.Sigal et al.(L.Sigal and M.Black.Measure
Locally, Reason Globally：Occlusion-sensitive articulated pose estimation.IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), 2006.) Bayesian frame is proposed in this article, the framework is filtered comprising sequential importance sampling and Annealed Particle, and has used multi-motion model in tracking.In order that 3 d pose recovers to more conform to anatomic joint constraint, while make search space dimensionality reduction, the framework from training data learning motion model, using the Euclidean distance difference of virtual tag as error in measurement.The shortcoming of the method is to extract accurate characteristics of image to require a great deal of time, and video tracking is limited by with the presence or absence of learning database, if not existing learning database, cannot complete video tracking.
Second, the human body motion tracking method based on model：The method does not need learning database, image information is directly extracted on target video image, the similarity function of target image and model is set up, then similarity function is optimized so as to search for optimal state in the state space of higher-dimension, so as to obtain accurate human body attitude.If the C.Sminchisescu and A.Jepson. of institut national de recherche en infomatique et automatique (INRIA) are in (C.Sminchisescu and A.Jepson.Generative Modeling for Continuous Non-Linearly
Embedded Visual Inference.International Conference on
Machine Learning (ICML), 2004) article in adopt this method the motion tracking realized using various manikins.Deutscher et
Al. in (J.Deutscher
and I.Reid.Articulated body motion capture by
Stochastic search.International Journal of ComputerVision (IJCV), 61 (2)：185-205,2004.) article in use border and silhouette as characteristics of image build weighting similarity function, using Annealed Particle filtering realize mans motion simulation.A similarity function is only set up due to the method, and the method for being used to optimize similarity function is easily ensnared into local optimum when optimal result is searched for, and causes the human body attitude for tracing into inaccurate, and also the time complexity of algorithm is high.
Hunan University's identified patent applications number 200910043537.5, " multi-human body tracking method based on attribute relational graph appearance model " of publication number CN101561928, the patent sets up attribute relational graph appearance model to present frame human testing region first, calculate the similarity with the attribute relational graph appearance model of previous frame tracking human body, the matching relationship of interframe human body is determined according to similarity, so that it is determined that human body tracking situation and acquisition movement locus.The deficiency that the method for the patent application publication is present is：Human body tracking can only be carried out to the sporter of fixed scene, only be not enough to accurately track human body attitude by the similarity of display model.
The content of the invention
It is an object of the invention to be directed to above-mentioned deficiency of the prior art, a kind of human body motion tracking method of the multiple-objection optimization based on model is proposed, realization carries out accurate human body attitude tracking to the sporter in different scenes.
Realizing the technical scheme of the object of the invention is, using the method based on model, set up human skeleton model, position and the half-tone information of artis are extracted using video image, build two Distance conformability degree functions, optimized by the multiple target Distance conformability degree function for building, under the constraint of skeleton length, the tracking to human motion attitude is realized using multi-objective optimization algorithm to the two Distance conformability degree functions.Implementation step includes as follows：
(1) three-dimensional human skeleton model is set up with the abstract method of bone：Will human skeleton be divided into 14 parts according to 15 joints, per part by a shaft-like bone model tormulation, straightway between space has the artis of three-dimensional coordinate with 14 represents this 14 shaft-like skeleton models, connect corresponding body joint point coordinate and constitute whole three-dimensional human skeleton model, when the D coordinates value of corresponding 15 artis of one group of movement human of input, human skeleton model will simulate the athletic posture of 3 D human body；
(2) human body video image is pre-processed
Human body video image 2a) is input into, human body silhouette is separately won to obtain by background subtraction, extract human body contour outline, axis micronization processes are carried out to human body contour outline, form human skeleton line；
Head, belly, knee, pin node coordinate position 2b) are obtained along skeleton line search on human skeleton line, remaining human body body joint point coordinate position is gone out using particle filter predicted detection；
(3) the characteristics of image r of the second-generation strip wave conversion Bandlet2 of video image is extracted, as the input of double gauss process, using double gauss TGP algorithms, the 3-dimensional coordinate artis v ' of the i-th frame human body is predictedi, i ∈ [1, N], obtain video sequence 3D artis be output as V ',
R=(r1, r2, r3..., rN)T, V '=(v1 ', v2 ', v3 ' ..., vN ') T ,]]>
Wherein, riFor the Bandlet2 of the i-th two field picture is characteristics of image, i ∈ [1, N], ()TRepresenting matrix turns order；
(4) human skeleton model is initialized
4a) to step 2b) the initial time video image artis position that obtains carries out manual demarcation, and setting the corresponding human skeleton of initial time human body attitude by nominal data is designated as v0, wherein v0Be 2b) in the human joint pointses position of the first frame video image that detects；
The t-1 moment 4b) is tracked the human skeleton that obtains as the initialization human skeleton of t, t ＞ 0；
(5) similarity function is built
5a) the 3D artis of human body is represented with V, 2D artis VqRepresent, VqIt is projections of the V in 2D planes, V is amount to be estimated：
V=(v1, v2, v3..., vN)T, Vq=(v1q, v2q, v3q ..., vNq) ,]]>
Wherein, viIt is the 3D artis of the i-th two field picture, i ∈ [1, N] are the 2D artis of the i-th frame, i ∈ [1, N], and N is video frame number；
The the i-th frame human body 3D artis V ' that will 5b) be predicted with double gauss TGP methods are projected in 2D planes, obtain the body joint point coordinate V ' of 2D projectionsp：
Vp′=(vp1′,vp2′,vp3′,...,vpN′)T,]]>
Wherein, it is projection of the i-th frame artis 3D artis on 2D, i ∈ [1, N]；
The Distance conformability degree function f for 5c) setting up respectively under 3D1(vi, v 'i) and 2D under Distance conformability degree function
(6) non-dominant neighborhood immune algorithm is utilized, in t to two Distance conformability degree function f1(vi, v 'i), bone length constraint under optimize, obtain one group of human skeleton similar to real human body athletic posture of t；
(7) skeleton is obtained by the human skeleton that step (6) is obtained to each in t, the Euclidean distance of the human skeleton artis that the skeleton joint point is traced into the t-1 moment is calculated, the most accurate human skeleton that the minimum human skeleton of Euclidean distance is traced into as t is selected.
The present invention has advantages below compared with prior art：
1st, due to obtaining more accurate human joint pointses picture position present invention uses particle filter prediction human joint pointses, the method that artis position is obtained compared with prior art is simpler, and time complexity is lower.
2nd, method of the present invention due to combining the popular production in current human's tracking field and discriminate, establishes the Distance conformability degree function of 2D and 3D respectively, can preferably utilize video image information.
3rd, the non-dominant neighborhood immune algorithm optimization object function due to having used multi-objective Evolutionary Algorithm of the invention, more existing single object optimization human body tracing method can avoid being absorbed in local optimum, improve the accuracy of mans motion simulation.
Brief description of the drawings
Fig. 1 is general flow chart of the invention.
Fig. 2 is the human joint pointses detection sub-process figure in the present invention.
Fig. 3 is three-dimensional tracking result figure of the present invention to the emulation experiment of walking posture.
Fig. 4 is three-dimensional tracking result figure of the present invention to the emulation experiment of attitude of boxing.
Specific embodiment
The present invention will be further described below in conjunction with the accompanying drawings.
Reference picture 1, of the invention to implement step as follows：
Step 1, sets up human skeleton model.
According to anatomical knowledge, although human skeleton is influenceed by age and health and constantly changed, the composition of skeleton is constant, and human body is generally comprised：Shin bone, femur, hipbone, trunk, radius, humerus, clavicle, neck, head.The present invention is expressed as human body by 15 artis and 14 skeleton patterns constituted with shaft-like bone in this case.Straightway between Virtual Space has the artis of three-dimensional coordinate with 14 represents this 14 shaft-like skeleton models.
It is i ∈ [1,15] by the coordinate representation of each artis, n ∈ [1, N], N are human motion video frame number to be tracked；The bone length that n-th frame human skeleton is expressed as two neighboring artis is expressed as p, q ∈ [1,15], the restrictive condition of human skeleton model is thus obtained | | Livn| |=li, i=1, wherein 2 ..., LiIt is 3 × 15 matrixes, liIt is i-th length of bone, m is total bone number；
Constrained in above-mentioned bone | | Livn| |=li, i=1 under 2 ..., whole three-dimensional human skeleton model constitute by adjacent artis connection, and when one group of correspondingly human motion is input into during the D coordinates value of 15 artis, human skeleton model will simulate the 3 D human body attitude of motion.
Step 2, preprocessed video image.
Reference picture 2, this step is implemented as follows：
2a) extract the skeleton line of human body silhouette：
Human body video image 2a1) is input into, background image is obtained using least square intermediate value LMedS methods；
The background image of acquisition and human motion image 2a2) are done into pixel difference, background difference image is obtained；
The segmentation noise in background difference image 2a3) is removed using morphological method to the background difference image for obtaining, clearly human body silhouette is obtained；
Human body silhouette outline 2a4) is obtained using border following algorithm to the human body silhouette for obtaining, the axis line thinning human body silhouette of silhouette outline is extracted, the skeleton line of human body silhouette is obtained；
2b) in step 2a) in obtain human body silhouette skeleton line on along skeleton line search, obtain to the end, root, knee, the coordinate position of pin node：
2b1) using concentric circle template along skeleton line search, will fall into annulus human body silhouettes point it is most when the center of circle as head node；
It is root node 2b2) to choose human body silhouette center of gravity position, using the arithmetic mean of instantaneous value of owner side shadow point x coordinate value as root node x coordinate, using the arithmetic mean of instantaneous value of y-coordinate value as root node y-coordinate；
2b3) three-dimensional human skeleton model is projected on the video images on the basis of root node, trunk central point, clavicle joint point and left and right buttocks artis is obtained；
2b4) according to head achieved above, root artis, detected using particle filter sell, the coordinate position of elbow, shoulder, knee and foot joint point.
Step 3：Extract the characteristics of image r of the second-generation strip wave conversion Bandlet2 of video image：
Pending video image 3a) is input into, human body block diagram in image is extracted, two-dimentional multi-scale wavelet transformation is carried out to block diagram；
3b) the image quad-tree partition algorithm after two-dimentional multi-scale wavelet transformation and bottom-up fusion rule are found and quantifies optimal geometry flow direction；
3c) by quantization after optimal geometry flow direction signal do one-dimensional wavelet transform, be reassembled as two dimensional form, obtain Bandelet2 coefficient matrixes；
3d) extract Bandlet2 feature r, r=(r of the maximum geometry statistical flow characteristic as image1... ri... rN)T, wherein, riIt is the Bandlet2 characteristics of image of the i-th two field picture, i ∈ [1, N], N are video frame number ()TThe transposition of representing matrix.
Step 4：The second-generation strip wave conversion Bandlet2 characteristics of image r of the video image extracted with step 3, as the input of double gauss method, predict the 3-dimensional coordinate artis v ' of the i-th frame human bodyi：
((V′)(d))T(vi′)(d)∝NR(0,KRKRr(KRr)TKR(r,r)),]]>
Wherein, NR() represents Gaussian process, ()TThe transposition of representing matrix, r is the bandlet2 features of input, and V ' is exported for the 3D artis of human body attitude to be predicted, V '=(v '1, v '2, v '3..., v 'N)T, N is video frame number, ((V ')(d))TIt is the d rows i.e. human body attitude of d frames of human body attitude V ' to be predicted, (v 'i)(d)It is the 3-dimensional artis v ' of the i-th frame human body to be predictediIn d-th body joint point coordinate, KR(r, r) is zero, KRIt is a matrix of N × N, KRIn the i-th row jth row element be (KR)ij, be a column vector for N × 1, in the i-th row element be
(KRr)i=KR(ri,r),]]>KR(ri, r)=cov (f (ri), f (r)),
(KR)ij=KR(ri, rj), KR(ri, rj)=cov (f (ri), f (rj)),
Cov (f (r in formulai), f (rj)) it is f (ri), f (rj) between covariance function, f (ri) be the i-th frame bandlet2 features zero-mean gaussian function, f (rj) be jth frame bandlet2 features zero-mean gaussian function, f (r) be input bandlet2 features zero-mean gaussian function.
Step 5：Initialization human skeleton model
5a) to step 2b) the initial time video image artis position that obtains carries out manual demarcation, and it is v to set the corresponding human skeleton of initial time human body attitude by nominal data0, wherein v0Be 2b) in the human joint pointses position of the first frame video image that detects；
The t-1 moment 5b) is tracked the human skeleton that obtains as the initialization human skeleton of t, t ＞ 0.
Step 6：Set up similarity function
The 3D artis for 6a) being obtained according to double gauss prediction and human joint pointses to be predicted, the 3D Distance conformability degree functions f set up under n-th frame video1(vn, v 'n)：
f1(vn,vn′)=Σi=115||vni-vni′||2,n∈[1,N],]]>
Wherein, N is video frame number, | | | |22 norms are represented, is artis to be predicted, be the artis that double gauss is predicted；
Projection of the artis and human joint pointses to be predicted for 6b) being predicted according to double gauss in 2D planes, the 2D Distance conformability degree functions set up under n-th frame video
f2(vnq,vpn′)=Σi=115||vniq-vpni′||2,n∈[1,N],]]>
Wherein, N is video frame number, | | | |22 norms are represented, is the projection of artis to be predicted, be the projection of the artis that double gauss is predicted.
Step 7：Optimization similarity function
Under the constraint of the skeleton length in two similarity functions and step 1 obtaining in step 6, setting solves two similarity function f1(vn, v 'n) and minimum value equation group：
argminf1(vn,vn′)=Σi=115||vni-vni′||2,argminf2(vnq,vpn′)=Σi=115||vniq-vpni′||2,s.t.||Livn||=li,i=1,2,···]]>
Wherein, liIt is i-th skeleton length, m is skeleton number, and n ∈ [1, N], N are video frame number, and arg min () is represented and minimized, | | | |2Represent 2 norms.
Using non-dominant neighborhood immune algorithm, the minimum value of equation group is solved under bone length constraint in t, obtain one group of human skeleton similar to real human body athletic posture of t.
Step 8：Selection human body optimal movement attitude
In the human skeleton that t is obtained to each by step 7, the Euclidean distance of the human skeleton artis that the skeleton joint point is traced into the t-1 moment is calculated, select the most accurate human skeleton that the minimum human skeleton of Euclidean distance is traced into as t.
Experiment simulation
Effect of the invention can be verified by following emulation experiment：
Emulation experiment of the invention is in Matlab
Compiled on 2010a and completed, performing environment is the HP work stations under Windows frameworks.HumanEva database of the video image from Brown Univ USA used by emulation experiment of the present invention, video image size is 320 × 240.
Emulation content
Walking states are tracked, as a result as shown in Figure 3 by emulation 1 using the present invention.Human body in Fig. 3 is raw video image, and the skeleton line of human body surface is the optimal motion state that tracking is obtained.
From figure 3, it can be seen that the attitude of ambiguity does not occur in tracking result, human motion attitude is accurately recovered, has shown that the present invention can realize accurately tracking to simple athletic posture.
Emulation 2, is tracked, as a result as shown in Figure 4 using the present invention to boxing state.Human body image in Fig. 4 is raw video image, and the skeleton line on human body image surface is the optimal motion state that tracking is obtained.
Figure 4, it is seen that tracking result does not have ambiguity attitude to occur, human motion attitude is accurately recovered, has shown that this method can also realize accurate tracking to complicated human motion state.
Analysis of simulation result：It is also seen that the present invention is essentially identical with real human motion attitude to different motion state video image tracking results from Fig. 3, Fig. 4, effectively solves the ambiguity problem of mans motion simulation, improves the Stability and veracity of tracking.Main reason is that this method employs two similarity functions, preferably using video image information, human skeleton length constraint is added when two similarity functions are minimized, limit the appearance of ambiguity human body attitude.
Claims (7)
1. the human body motion tracking method that a kind of production and discriminate combine, comprises the following steps：
Three-dimensional human skeleton model is set up with the abstract method of bone：Will human skeleton be divided into 14 parts according to 15 joints, per part by a shaft-like bone model tormulation, straightway between space has the artis of three-dimensional coordinate with 14 represents this 14 shaft-like skeleton models, connect corresponding body joint point coordinate and constitute whole three-dimensional human skeleton model, when the D coordinates value of corresponding 15 artis of one group of movement human of input, human skeleton model will simulate the athletic posture of 3 D human body；
Pretreatment human body video image：Input human body video image, human body silhouette is separately won to obtain by background subtraction, extracts human body contour outline, and axis micronization processes are carried out to human body contour outline, forms human skeleton line；Head, belly, knee, pin node coordinate position are obtained along skeleton line search on human skeleton line, remaining human body body joint point coordinate position is gone out using particle filter predicted detection；The characteristics of image r of the second-generation strip wave conversion Bandlet2 of video image is extracted, as the input of double gauss process, using double gauss TGP algorithms, the 3-dimensional coordinate artis v ' of the i-th frame human body is predictedi, i ∈ [1, N], obtain video sequence 3D artis be output as V ',
R=(r1, r2, r3..., rN)T, V '=(v1 ', v2 ', v3 ' ..., vN ') T ,]]>
Wherein, riIt is the Bandlet2 characteristics of image of the i-th two field picture, i ∈ [1, N], ()TRepresenting matrix turns order；
Initialization human skeleton model
To step 2b) the initial time video image artis position that obtains carries out manual demarcation, and setting the corresponding human skeleton of initial time human body attitude by nominal data is designated as v0, wherein v0Be 2b) in the human joint pointses position of the first frame video image that detects；The t-1 moment is tracked the human skeleton that obtains as the initialization human skeleton of t, t ＞ 0；
Build similarity function
The 3D artis of human body is represented with V, 2D artis VqRepresent, VqIt is projections of the V in 2D planes, V is amount to be estimated：
V=(v1, v2, v3..., vN)T, Vq=(v1q, v2q, v3q ..., vNq) ,]]>
Wherein, viIt is the 3D artis of the i-th two field picture, i ∈ [1, N] are the 2D artis of the i-th frame, i ∈ [1, N], and N is video frame number；
The the i-th frame human body 3D artis V ' that will be predicted with double gauss TGP methods are projected in 2D planes, obtain the body joint point coordinate V ' of 2D projectionsp：
Vp′=(vp1′,vp2′,vp3′,...,vpN′)T,]]>
Wherein, it is projection of the i-th frame artis 3D artis on 2D, i ∈ [1, N]；
The Distance conformability degree function f set up respectively under 3D1(vi, v 'i) and 2D under Distance conformability degree function
Using non-dominant neighborhood immune algorithm, to two Distance conformability degree function f of t1(vi, v 'i), under bone length constraint solve minimum value, obtain one group of human skeleton similar to real human body athletic posture of t；Skeleton is obtained by the human skeleton that step (6) is obtained to each in t, the Euclidean distance of the human skeleton artis that the skeleton joint point is traced into the t-1 moment is calculated, the most accurate human skeleton that the minimum human skeleton of Euclidean distance is traced into as t is selected.
2. human body motion tracking method according to claim 1, the use double gauss TGP algorithms wherein described in step (3), predict v 'i, carry out as follows：
((V′)(d))T(vi′)(d)∝NR(0,KRKRr(KRr)TKR(r,r))]]>
Wherein, NR() represents Gaussian process, and r is the bandlet2 features of input, ((V ')(d))TIt is the d rows i.e. human body attitude of d frames of human body attitude V ' to be predicted, (v 'i)(d)It is the 3-dimensional artis v of the i-th frame human body to be predictedi' in d-th body joint point coordinate, KR(r, r) is zero, KRIt is a matrix of N × N, KRIn the i-th row jth row element be (KR)ij, be a column vector for N × 1, in the i-th row element be
(KRr)i=KR(ri,r),]]>KR(ri, r)=cov (f (ri), f (r)),
(KR)ij=KR(ri, rj), KR(ri, rj)=cov (f (ri), f (rj)),
Cov (f (r in formulai), f (rj)) it is f (ri), f (rj) between covariance function, f (ri) be the i-th frame bandlet2 features zero-mean gaussian function, f (rj) be jth frame bandlet2 features zero-mean gaussian function, f (r) be input bandlet2 features zero-mean gaussian function.
3. human body motion tracking method according to claim 1, wherein step 5c) in the norm of 3D distances 2 be：
f1(vi,vi′)=Σi=115||vni-vni′||2,]]>
Wherein, it is n-th frame image human skeleton artis v to be estimatednIn i-th artis 3D coordinates, be n-th frame image human skeleton artis v ' that double gauss is predictednIn i-th artis 3D coordinates, n ∈ [1, N].
4. human body motion tracking method according to claim 1, wherein step 5c) in the norm of 2D distances 2 be：
f2(viq,vpi′)=Σi=115||vniq-vpni′||2,]]>
Wherein, it is n-th frame image human skeleton artis v to be estimatednThe 2D coordinates of i-th artis projected in 2D planes, are to predict n-th frame image framework artis v ' using double gaussnThe 2D coordinates of i-th artis projected in 2D planes, n ∈ [1, N].
5. human body motion tracking method according to claim 1, wherein step 2a) described in extraction human body contour outline, axis micronization processes are carried out to human body contour outline, form human skeleton line, it is that background image is first obtained using least square intermediate value LMedS methods, human motion image and background image are done into pixel difference, background difference image is obtained；The segmentation noise in background difference image is removed using morphological method again, clearly human body silhouette is obtained；Human body silhouette outline is finally obtained using border following algorithm, the axis line thinning human body silhouette of silhouettes is extracted, the skeleton line of human body silhouette is obtained.
6. human body motion tracking method according to claim 1, wherein step 3) described in extraction video image second-generation strip wave conversion Bandlet2 characteristics of image, carry out in accordance with the following steps：
Pending video image is first input into, human body block diagram in image is extracted, two-dimentional multi-scale wavelet transformation is carried out to block diagram；
Optimal geometry flow direction is found and quantified to the image quad-tree partition algorithm after two-dimentional multi-scale wavelet transformation and bottom-up fusion rule；
Optimal geometry flow direction signal after by quantization does one-dimensional wavelet transform, is reassembled as two dimensional form, obtains Bandelet2 coefficient matrixes；
Maximum geometry statistical flow characteristic is extracted as final image character representation.
7. human body motion tracking method according to claim 1, wherein step 6) described in utilization non-dominant neighborhood immune algorithm, to two Distance conformability degree function f of t1(vn, v 'n), bone length constraint under solve minimum value, carried out according to equation below：
argminf1(vn,vn′)=Σi=115||vni-vni′||2,argminf2(vnq,vpn′)=Σi=115||vniq-vpni′||2,s.t.||Livn||=li,i=1,2,···]]>
Wherein, liIt is i-th skeleton length, m is skeleton number, and n ∈ [1, N], N are video frame number, and arg min () is represented and minimized, | | | |2Represent 2 norms.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510870921.8A CN106815855A (en) | 2015-12-02 | 2015-12-02 | Based on the human body motion tracking method that production and discriminate combine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510870921.8A CN106815855A (en) | 2015-12-02 | 2015-12-02 | Based on the human body motion tracking method that production and discriminate combine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106815855A true CN106815855A (en) | 2017-06-09 |
Family
ID=59109132
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510870921.8A Pending CN106815855A (en) | 2015-12-02 | 2015-12-02 | Based on the human body motion tracking method that production and discriminate combine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106815855A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416258A (en) * | 2018-01-23 | 2018-08-17 | 华侨大学 | A kind of multi-human body tracking method based on human body model |
| WO2019192205A1 (en) * | 2018-04-02 | 2019-10-10 | 京东方科技集团股份有限公司 | Method and apparatus for identifying representation information of limb in image, device, and computer-readable storage medium |
| CN110477921A (en) * | 2019-08-15 | 2019-11-22 | 合肥工业大学 | The height measurement method returned based on skeleton broken line Ridge |
| CN110728739A (en) * | 2019-09-30 | 2020-01-24 | 杭州师范大学 | Virtual human control and interaction method based on video stream |
| CN111524183A (en) * | 2020-04-07 | 2020-08-11 | 上海交通大学 | Target row and column positioning method based on perspective projection transformation |
| CN111539981A (en) * | 2020-04-13 | 2020-08-14 | 北京航空航天大学 | Motion prediction system based on artificial intelligence |
| CN112183316A (en) * | 2020-09-27 | 2021-01-05 | 中山大学 | Method for measuring human body posture of athlete |
| CN113158942A (en) * | 2021-04-29 | 2021-07-23 | 泽恩科技有限公司 | Segmentation algorithm and device for detecting motion human behavior |
| CN113673494A (en) * | 2021-10-25 | 2021-11-19 | 青岛根尖智能科技有限公司 | Human body posture standard motion behavior matching method and system |
-
2015
- 2015-12-02 CN CN201510870921.8A patent/CN106815855A/en active Pending
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416258B (en) * | 2018-01-23 | 2020-05-08 | 华侨大学 | Multi-human body tracking method based on human body part model |
| CN108416258A (en) * | 2018-01-23 | 2018-08-17 | 华侨大学 | A kind of multi-human body tracking method based on human body model |
| US11354925B2 (en) | 2018-04-02 | 2022-06-07 | Beijing Boe Optoelectronics Technology Co., Ltd. | Method, apparatus and device for identifying body representation information in image, and computer readable storage medium |
| WO2019192205A1 (en) * | 2018-04-02 | 2019-10-10 | 京东方科技集团股份有限公司 | Method and apparatus for identifying representation information of limb in image, device, and computer-readable storage medium |
| CN110477921A (en) * | 2019-08-15 | 2019-11-22 | 合肥工业大学 | The height measurement method returned based on skeleton broken line Ridge |
| CN110728739A (en) * | 2019-09-30 | 2020-01-24 | 杭州师范大学 | Virtual human control and interaction method based on video stream |
| CN110728739B (en) * | 2019-09-30 | 2023-04-14 | 杭州师范大学 | Virtual human control and interaction method based on video stream |
| CN111524183A (en) * | 2020-04-07 | 2020-08-11 | 上海交通大学 | Target row and column positioning method based on perspective projection transformation |
| CN111539981A (en) * | 2020-04-13 | 2020-08-14 | 北京航空航天大学 | Motion prediction system based on artificial intelligence |
| CN111539981B (en) * | 2020-04-13 | 2023-03-10 | 北京航空航天大学 | Motion prediction system based on artificial intelligence |
| CN112183316A (en) * | 2020-09-27 | 2021-01-05 | 中山大学 | Method for measuring human body posture of athlete |
| CN112183316B (en) * | 2020-09-27 | 2023-06-30 | 中山大学 | Athlete human body posture measuring method |
| CN113158942A (en) * | 2021-04-29 | 2021-07-23 | 泽恩科技有限公司 | Segmentation algorithm and device for detecting motion human behavior |
| CN113673494A (en) * | 2021-10-25 | 2021-11-19 | 青岛根尖智能科技有限公司 | Human body posture standard motion behavior matching method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135375B (en) | Multi-person attitude estimation method based on global information integration | |
| Kamal et al. | A hybrid feature extraction approach for human detection, tracking and activity recognition using depth sensors | |
| CN106815855A (en) | Based on the human body motion tracking method that production and discriminate combine | |
| CN107423730B (en) | Human gait behavior active detection and recognition system and method based on semantic folding | |
| Zhang et al. | Martial arts, dancing and sports dataset: A challenging stereo and multi-view dataset for 3d human pose estimation | |
| CN102184541B (en) | Multi-objective optimized human body motion tracking method | |
| CN102682452A (en) | Human movement tracking method based on combination of production and discriminant | |
| Uddin et al. | Human activity recognition using body joint‐angle features and hidden Markov model | |
| CN110472604B (en) | Pedestrian and crowd behavior identification method based on video | |
| CN102622766A (en) | Multi-objective optimization multi-lens human motion tracking method | |
| CN102074034B (en) | Multi-model human motion tracking method | |
| CN109299659A (en) | A kind of human posture recognition method and system based on RGB camera and deep learning | |
| Zhang et al. | Unsupervised skeleton extraction and motion capture from 3D deformable matching | |
| CN114067358A (en) | Human body posture recognition method and system based on key point detection technology | |
| CN112131908A (en) | Action identification method and device based on double-flow network, storage medium and equipment | |
| Zhou et al. | Learning to estimate 3d human pose from point cloud | |
| JP2019096113A (en) | Processing device, method and program relating to keypoint data | |
| CN109919137B (en) | Pedestrian structural feature expression method | |
| Chang et al. | The model-based human body motion analysis system | |
| Uddin et al. | Human Activity Recognition via 3-D joint angle features and Hidden Markov models | |
| CN102663779A (en) | Human motion tracking method based on stochastic Gaussian hidden variables | |
| CN109657634A (en) | A kind of 3D gesture identification method and system based on depth convolutional neural networks | |
| Liu | Aerobics posture recognition based on neural network and sensors | |
| Lin et al. | Overview of 3d human pose estimation | |
| Liu et al. | Key algorithm for human motion recognition in virtual reality video sequences based on hidden markov model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170609 |