CN103021009B - Motion diagram transition point selecting method based on nonlinear manifold learning - Google Patents
Motion diagram transition point selecting method based on nonlinear manifold learning Download PDFInfo
- Publication number
- CN103021009B CN103021009B CN201210554973.0A CN201210554973A CN103021009B CN 103021009 B CN103021009 B CN 103021009B CN 201210554973 A CN201210554973 A CN 201210554973A CN 103021009 B CN103021009 B CN 103021009B
- Authority
- CN
- China
- Prior art keywords
- data
- frame
- motion
- critical
- calculate
- 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
Landscapes
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a motion diagram transition point selecting method based on nonlinearity manifold learning and belongs to the technical field of computer image processing. The method includes the steps of dimensionality reduction analysis of high dimensional data, extraction of key data segments, calculation of key data segment interframe similarity and construction of motion diagrams.
Description
Technical field
The present invention relates to a kind of motion diagram transition point choosing method based on non-linearity manifold study, belong to computer image processing technology field.
Background technology
In recent years, along with the progress of computer hardware technique, Computer Animated Graph obtains development at full speed, computer animation refers to the treatment technology adopting graph and image, based on solid modelling and Realistic representation technology, generate a series of scenery picture by means of programming or animation soft.It relates to the various fields such as image processing techniques, motion control principle, video technique and art, progressively becomes the field of a multiple subject and technological synthesis with the feature of uniqueness.Wherein, the development of movement capturing technology, the more diversified and motion of virtual human of complexity of the data genaration that people can utilize capture device to grab.But because motion capture device is expensive and be limited to the external condition of seizure, all can not catch required human body movement data, motion data reuse technology and the human body animation synthetic technology based on motion diagram produce thus at every turn.The human body movement data of catching is preserved according to type of sports classification and builds motion diagram, when synthesizing new human body movement data, only need coverage motion figure and in conjunction with interpolation technique can by existing human body movement data fragment synthesize new needed for human body movement data.Therefore, the motion synthetic technology research based on movement capturing data is one of Computer Animated Graph important field of research.
After the human motion synthetic method based on motion diagram in 2002 produces, in human body movement data synthesis field, the human motion synthetic method based on motion diagram has become current main method.As shown in Figure 2, the basic skills of motion diagram is by type of sports segmentation by movement capturing data, every frame is a node, then each internodal similarity is calculated, limit is configured between the node of threshold value set by meeting, finally form motion diagram, the process of carrying out human motion synthesis is exactly the process of searching for desired path on motion diagram.Although through development for many years, still there is a lot of problem in it, such as the definition on motion diagram mid point, limit, choosing of transition point, the planning etc. of searching route on motion diagram.
On the other hand, method based on non-linearity manifold study is introduced in human motion synthesis field, non-linearity manifold study can carry out Dimension Reduction Analysis to higher-dimension human body movement data, conventional non-linearity manifold study method has ISOMAP, ST-ISOMAP, SOM and LLE etc., first three kind belongs to global approach, and latter belongs to partial approach.ISOMAP dimension reduction method can be used for the segmentation of motor segment, extracts border key frame, to distinguish segmentation dissimilar in the original motion data section.Human body movement data fragment can be projected to the enterprising rearrangement of low dimensional manifold and obtain new exercise data fragment by ST-ISOMAP method.And SOM method obtains the distribution of high dimensional data on low dimensional manifold.
Existing research shows, method based on overall non-linearity manifold study is suitable for the operation such as segmentation, extraction key frame of human motion fragment, low-dimensional data after dimensionality reduction can well reflect the higher-dimension attitude of motion sequence, effectively can excavate the most essential motion feature of exercise data fragment.Meanwhile, reduce motion diagram transition point complexity access time and remain a challenging task.
Summary of the invention
The present invention is directed to the proposition of above problem, and develop a kind of motion diagram transition point choosing method based on non-linearity manifold study, the method is by setting up the similarity calculated between critical movements data segment, solve emphatically jump-point complexity access time in motion diagram building process high, choose inaccurate problem, thus improve the structure efficiency of motion diagram, make the more level and smooth nature of exercise data generated.
The present invention includes following steps:
Step one: the Dimension Reduction Analysis of high dimensional data.
Step 2: extract critical data segment.
Step 3: calculate critical data segment frame-to-frame coherence.
Step 4: build motion diagram.
Principle of the present invention: carry out dimension-reduction treatment to higher-dimension human body movement data by ISOMAP dimension-reduction algorithm, uses the low-dimensional data after dimensionality reduction, draws low-dimensional characteristic curve.The critical data segment in human motion fragment is chosen according to low-dimensional characteristic curve, calculate the similarity of critical movements data segment interframe, the frame meeting set threshold value connects into limit, the motion diagram structure that final formation one is made up of critical data segment, when synthesizing new human body movement data, only need travel through this motion diagram.
The present invention compared with prior art has the following advantages:
In Table 1, method A is the standard method calculating transition point.Method B is a kind of transition point fast selecting algorithm.Visible by table one, the distance matrix calculating gained is respectively A [344*163], B [115*55], C [250*88].Time loss correlation data represents the time calculated needed for distance map, method A is 117.572319 seconds, and method B is 19.229195 seconds, and institute extracting method C of the present invention is 45.487492 seconds, therefore institute of the present invention extracting method is better than canonical algorithm in time efficiency, close with express method.Section 3 correlation data accuracy rate represent transition point that current algorithm finds account for all can the ratio of tie point, because method A is canonical algorithm, calculate the similarity of all interframe of current kinetic sequence, therefore the accuracy rate of assumption method A is 100%, setting threshold value M, other 2 kinds of methods all A are that substrate calculates accuracy rate, and represent accuracy rate with R, what current algorithm found transition point adds up to S
i(i=a, b, c), owing to considering that canonical algorithm can produce a large amount of invalid edges when calculating, particularly when the 2 sections of type of sports participating in calculating are similar, do not meet physical law 2 frames motion between also may connect into limit, therefore set a weights ρ, ρ value is different according to the difference of 2 sections of type of sports of participation calculating, in 0 ~ 1 scope, be worth larger, type of sports gap is larger.Then computing formula is:
R＝ρ(S
i/S
a)，
By formulae discovery, method B accuracy rate is 33.4%, and institute extracting method C accuracy rate is 78.1% herein.Contrast visible, the accuracy rate that institute of the present invention extracting method finds suitable transition point is better than quick back-and-forth method, and be near the mark algorithm.Therefore, motion diagram transition point choosing method based on non-linearity manifold study has certain advantage on time and accuracy rate, this method, while significantly reducing time loss, maintains certain accuracy rate, is a kind of motion diagram transition point Algorithms of Selecting effectively reliably.
Table one: contrast with additive method
| Method | A | B | C |
| Frame number | 344*163 | 115*55 | 250*88 |
| Time loss | 117.572319s | 19.229195s | 45.487492s |
| Accuracy rate | 100％ | 33.4％ | 78.1％ |
Accompanying drawing explanation
Fig. 1 algorithm flow chart of the present invention.
Fig. 2 standard movement figure Computing Principle.
Fig. 3 human body movement data low-dimensional characteristic curve schematic diagram.
Fig. 4 critical data segment chooses schematic diagram 1.
Fig. 5 critical data segment chooses schematic diagram 2.
Fig. 6 Similarity measures result figure.In figure, color gets over depths, represents that distance is more close, can set up the point of motion diagram fillet.
Embodiment
Below in conjunction with accompanying drawing, the present invention will be further explained, and unit of the present invention embodiment is:
Embodiments of the invention are implemented under premised on technical solution of the present invention, give detailed embodiment and concrete operating process, but protection scope of the present invention are not limited to following embodiment.
Be illustrated in figure 1 algorithm flow chart of the present invention, it specifically comprises following sport technique segment:
Step one: the Dimension Reduction Analysis of high dimensional data
Use ISOMAP non-linearity manifold study algorithm, dimension-reduction treatment is carried out to higher-dimension human body movement data, obtains the low dimensional manifold structure of original motion sequence, according to the difference of type of sports, draw the low-dimensional characteristic curve matched.
(1) get n dimension human body movement data, then input data set channel=[L × n] that a segment length is L frame, according to CMU database data used herein, each frame data comprises n=96 dimension data;
(2) Neighbor Points of each point is calculated, with k nearest neighbor or epsilon neighborhood;
(3) on sample set, define a non-directed graph, build non-directed graph G with the data point of local neighborhood, while represent the connection within the scope of local neighborhood between point;
(4) calculate the bee-line of non-directed graph, use the bee-line of 2 points (i, j) in dijkstra's algorithm calculating chart, the distance matrix of gained is D
g:
D
G＝{d
G(i，j)}；
(5) use MDS to solve low-dimensional and embed stream shape;
As shown in Figure 3, Fig. 3 represents the low-dimensional characteristic curve of one section of three-dimensional motion after ISOMAP dimensionality reduction to method disclosed by the invention.Curve is with the change of the human body higher-dimension characteristics of motion, and in the turning point of curve, the human motion of higher dimensional space changes its original motion state.
Step 2: extract critical data segment
According to human motion low-dimensional characteristic curve, extract critical movements section, in this critical movements section, calculate frame-to-frame coherence.
Based on the consideration to transition smoother, adopt a length to be the window of W, the data in window are added new data set Y
i, the value of W is determined by formula:
W＝L/(ρC)；
Wherein, L is frame length, and ρ is weights, depends on that the frequency that human body attitude changes, C are the quantity on characteristic curve summit.After determining the value of W, the frame number in record window, extracts corresponding data in data set channel, stored in new data set Y
iin, i is corresponding motion sequence sequence number;
As shown in Figure 4: gray area is in one section of human motion, the critical data segment extracted is needed.
Step 3: calculate critical data segment frame-to-frame coherence
Use the computing method of Euclidean distance, calculate the similarity of the intersegmental every frame of critical data, draw distance map.
First, the critical movements data segment extracted in step 2 is set up data set, the critical section extracted in former exercise data is a data set;
Secondly, calculate frame-to-frame coherence, calculate the distance of each interframe between two, adopt Euclidean distance to represent.By meeting 2 connections of set threshold value, form a jump-point;
Accompanying drawing 6 is utilize institute herein to carry algorithm walk one group and run the distance map finally calculating similarity through dimensionality reduction, extraction critical movements data segment, and in figure, color gets over depths, represents apart from more close, can set up the point of motion diagram fillet.Table two is depicted as after dimension-reduction treatment, and critical data segment frame number and the raw data section frame number of extraction contrast.Use institute's extracting method herein, compare the standard method of extracting transition point, extraction rate improves more than 40%, extracts percentage of head rice and is about 78% ~ 100%.
As shown in Figure 6: use two dimensional gray figure to represent the size of distance value, distance is nearer, and color is darker, and expression can set up jump-point.Otherwise color is more shallow, means that 2 frame motion difference are excessive, can not jump-point be set up.
Step 4: build motion diagram
Set a threshold value M, all interframe being less than threshold value M is connected into limit, after cutting a process, be built into motion diagram.Fig. 2 is the Computing Principle of standard movement figure, and left side be exercise data section as shown in Figure 2, after right side represents calculate frame-to-frame coherence in exercise data section, can set up redirect limit between the frame of threshold value set by meeting.
Example is chosen one group and is walked and run motion and verify, exercise data low-dimensional characteristic curve as shown in Figure 3, critical data segment is chosen as shown in accompanying drawing 4 and accompanying drawing 5, in the diagram, gray area is in one section of human motion, need the critical data segment extracted, corresponding to the grey in Fig. 5 or black color dots, a point represents frame data.
Accompanying drawing 6 is utilize institute herein to carry algorithm walk one group and run the distance map finally calculating similarity through dimensionality reduction, extraction critical movements data segment, and in figure, color gets over depths, represents apart from more close, can set up the point of motion diagram fillet.Table two is depicted as after dimension-reduction treatment, and critical data segment frame number and the raw data section frame number of extraction contrast.Use institute's extracting method herein, compare the standard method of extracting transition point, extraction rate improves more than 40%, extracts percentage of head rice and is about 78% ~ 100%.
Table two: critical data segment extracts result
| Exercise data | Original frame number | Frame number after extracting |
| Walk | 344 | 150 |
| Run | 174 | 110 |
| Jump | 484 | 225 |
| Dancing | 1124 | 570 |
| Play football | 360 | 250 |
The above; be only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to according to technical scheme of the present invention and inventive concept thereof and replace or change, all should be encompassed within protection scope of the present invention.
Claims (1)
1., based on a motion diagram transition point choosing method for non-linearity manifold study, it is characterized in that: comprise the following steps:
Step one: the Dimension Reduction Analysis of high dimensional data
Use ISOMAP non-linearity manifold study algorithm, dimension-reduction treatment is carried out to higher-dimension human body movement data, obtains the low dimensional manifold structure of original motion sequence, according to the difference of type of sports, draw the low-dimensional characteristic curve matched;
(1) get n dimension human body movement data, then input data set channel=[L × n] that a segment length is L frame, according to CMU database data used herein, each frame data comprises n=96 dimension data;
(2) Neighbor Points of each point is calculated, with k nearest neighbor or epsilon neighborhood;
(3) on sample set, define a non-directed graph, build non-directed graph G with the data point of local neighborhood, while represent the connection within the scope of local neighborhood between point;
(4) calculate the bee-line of non-directed graph, use the bee-line of 2 points (i, j) in dijkstra's algorithm calculating chart, the distance matrix of gained is D
g:
D
G＝{d
G(i，j)}；
(5) use MDS to solve low-dimensional and embed stream shape;
Step 2: extract critical data segment
According to human motion low-dimensional characteristic curve, extract critical movements section, in this critical movements section, calculate frame-to-frame coherence;
Based on the consideration to transition smoother, adopt a length to be the window of W, the data in window are added new data set Y
i, the value of W is determined by formula:
W＝L/(ρC)；
Wherein, L is frame length, and ρ is weights, depends on that the frequency that human body attitude changes, C are the quantity on characteristic curve summit; After determining the value of W, the frame number in record window, extracts corresponding data in data set channel, stored in new data set Y
iin, i is corresponding motion sequence sequence number;
Step 3: calculate critical data segment frame-to-frame coherence
Use the computing method of Euclidean distance, calculate the similarity of the intersegmental every frame of critical data, draw distance map;
First, the critical movements data segment extracted in step 2 is set up data set, the critical section extracted in former exercise data is a data set;
Secondly, calculate frame-to-frame coherence, calculate the distance of each interframe between two, adopt Euclidean distance to represent; By meeting 2 connections of set threshold value, form a jump-point;
Step 4: build motion diagram
Set a threshold value M, all interframe being less than threshold value M is connected into limit, after cutting a process, be built into motion diagram.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210554973.0A CN103021009B (en) | 2012-12-14 | 2012-12-14 | Motion diagram transition point selecting method based on nonlinear manifold learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210554973.0A CN103021009B (en) | 2012-12-14 | 2012-12-14 | Motion diagram transition point selecting method based on nonlinear manifold learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103021009A CN103021009A (en) | 2013-04-03 |
| CN103021009B true CN103021009B (en) | 2015-02-11 |
Family
ID=47969575
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210554973.0A Active CN103021009B (en) | 2012-12-14 | 2012-12-14 | Motion diagram transition point selecting method based on nonlinear manifold learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103021009B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504731B (en) * | 2014-12-19 | 2017-08-22 | 西安理工大学 | Human motion synthetic method based on motion diagram |
| CN104573194A (en) * | 2014-12-20 | 2015-04-29 | 西安工业大学 | Recognition method for subassembly in assembly sequence planning |
| CN112188233B (en) * | 2019-07-02 | 2022-04-19 | 北京新唐思创教育科技有限公司 | Method, device and equipment for generating spliced human body video |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100583158C (en) * | 2008-01-14 | 2010-01-20 | 浙江大学 | Cartoon animation fabrication method based on video extracting and reusing |
| CN102521843B (en) * | 2011-11-28 | 2014-06-04 | 大连大学 | Three-dimensional human body motion analysis and synthesis method based on manifold learning |
-
2012
- 2012-12-14 CN CN201210554973.0A patent/CN103021009B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN103021009A (en) | 2013-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110472554A (en) | Table tennis action identification method and system based on posture segmentation and crucial point feature | |
| D’Orazio et al. | A review of vision-based systems for soccer video analysis | |
| CN105847804B (en) | A kind of up-conversion method of video frame rate based on sparse redundant representation model | |
| Liu et al. | Human pose estimation in video via structured space learning and halfway temporal evaluation | |
| CN102289948B (en) | Multi-characteristic fusion multi-vehicle video tracking method under highway scene | |
| CN103942751B (en) | A kind of video key frame extracting method | |
| Chen et al. | Sportscap: Monocular 3d human motion capture and fine-grained understanding in challenging sports videos | |
| EP3665653A1 (en) | Techniques for rendering three-dimensional animated graphics from video | |
| 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 | |
| CN104461000B (en) | A kind of on-line continuous human motion identification method based on a small amount of deleted signal | |
| CN103227888B (en) | A kind of based on empirical mode decomposition with the video stabilization method of multiple interpretational criteria | |
| CN110473266A (en) | A kind of reservation source scene figure action video generation method based on posture guidance | |
| CN106127804A (en) | The method for tracking target of RGB D data cross-module formula feature learning based on sparse depth denoising own coding device | |
| CN102360504A (en) | Self-adaptation virtual and actual three-dimensional registration method based on multiple natural characteristics | |
| CN102682452A (en) | Human movement tracking method based on combination of production and discriminant | |
| CN103021009B (en) | Motion diagram transition point selecting method based on nonlinear manifold learning | |
| Fossati et al. | Bridging the gap between detection and tracking for 3D monocular video-based motion capture | |
| CN108200432A (en) | A kind of target following technology based on video compress domain | |
| Gafni et al. | Single-shot freestyle dance reenactment | |
| CN104866853A (en) | Method for extracting behavior characteristics of multiple athletes in football match video | |
| Zhu et al. | Tracking multiple objects through occlusion with online sampling and position estimation | |
| CN103456012B (en) | Based on visual human hand detecting and tracking method and the system of maximum stable area of curvature | |
| KR102467903B1 (en) | Method for presenting motion by mapping of skeleton employing Augmented Reality | |
| Kalafatić et al. | Multiple object tracking for football game analysis | |
| Xie et al. | Hierarchical dynamic programming module for human pose refinement |
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 |