CN101470898B - Automatic analysis method for synchronization of two-person synchronized diving - Google Patents
Automatic analysis method for synchronization of two-person synchronized diving Download PDFInfo
- Publication number
- CN101470898B CN101470898B CN2007103042260A CN200710304226A CN101470898B CN 101470898 B CN101470898 B CN 101470898B CN 2007103042260 A CN2007103042260 A CN 2007103042260A CN 200710304226 A CN200710304226 A CN 200710304226A CN 101470898 B CN101470898 B CN 101470898B
- Authority
- CN
- China
- Prior art keywords
- sportsman
- outline
- synchronism
- diving
- value
- 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.)
- Expired - Fee Related
Links
- 230000009189 diving Effects 0.000 title claims abstract description 56
- 230000001360 synchronised effect Effects 0.000 title claims description 86
- 238000004458 analytical method Methods 0.000 title description 6
- 238000011156 evaluation Methods 0.000 claims abstract description 104
- 238000000034 method Methods 0.000 claims abstract description 63
- 230000006870 function Effects 0.000 claims abstract description 62
- 239000000284 extract Substances 0.000 claims abstract description 26
- 230000009471 action Effects 0.000 claims description 89
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 13
- 239000004744 fabric Substances 0.000 claims description 12
- 238000010586 diagram Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 6
- VYMDGNCVAMGZFE-UHFFFAOYSA-N phenylbutazonum Chemical compound O=C1C(CCCC)C(=O)N(C=2C=CC=CC=2)N1C1=CC=CC=C1 VYMDGNCVAMGZFE-UHFFFAOYSA-N 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 7
- 238000012549 training Methods 0.000 description 6
- 230000000386 athletic effect Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000002386 leaching Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 241000209094 Oryza Species 0.000 description 2
- 235000007164 Oryza sativa Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 235000009566 rice Nutrition 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009182 swimming Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000003292 diminished effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- SBNFWQZLDJGRLK-UHFFFAOYSA-N phenothrin Chemical compound CC1(C)C(C=C(C)C)C1C(=O)OCC1=CC=CC(OC=2C=CC=CC=2)=C1 SBNFWQZLDJGRLK-UHFFFAOYSA-N 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Abstract
The invention provides a method for analyzing the synchronism of double-people diving, which mainly comprises: extracting movement targets on the basis of dynamic background reconstruction in movement videos of double-people diving, representing and extracting the synchronism characteristics on the basis of synchronism grading factors in diving rules, and adopting a bias character statistic learning method to construct synchronism evaluation functions to carry out synchronism evaluation. The method can automatically extract athlete external outlines in the movement videos of double-people diving. The method provides a method for effectively representing the synchronism characteristics according to the diving rules. The invention also introduces the sequencing idea which is normally used in bias character statistic learning into the problem of constructing the synchronism evaluation functions of the movement videos of double-people diving, and the absolute score problem is converted into the relative ordering problem. Finally, the synchronism of double-people diving can be automatically estimated through calculating the synchronism evaluation function value of the movement videos of double-people diving. The method can estimate the synchronism of double-people diving accurately, effectively and automatically.
Description
Technical field
The invention belongs to the Computer Applied Technology field, be specifically related to computer vision technique, relate in particular to the method that the analysis of synchronised diving content of multimedia synchronism is evaluated automatically.
Background technology
In order to reduce the unjustness in the match, The present computer technology is applied in the sports tournament more and more, and Britain Hawk-Eye company has developed one and overlapped the complementary judgment system that is applied to tennis tournament: the Hawk-Eye system, and obtained great success.The Motion Technology of Switzerland DartFish company exploitation and tactics video recording analytic system DartFish system can extract athletic technical parameter (for example, joint position and angle, barycenter or the like) from sports video.The DartFish system can provide corresponding guidance to the sportsman through analyzing these information, and then improves the level of training.
The synchronised diving competition is the sports items (medal sport is listed in the Sydney Olympic Games 2000 in) of just formally carrying out recent years, is the racing dive of being done action by two sportsmen simultaneously from springboard or diving tower take-off.Match not only will be evaluated two diver performance of technical movements separately, and will take all factors into consideration the sportsman cooperates aspects such as time, entry angle, relative distance and time at take-off, action in the air consistance and harmony.In the Olympic Games, World Championships and world cup synchronised diving competition, 9 responsible marking of judge (wherein 4 responsible assessment technology branches are responsible for evaluation branch synchronously for other 5).Synchronism judge's judgement is quite crucial for match.And because shorter 1s-2s of dive duration is again the evaluation of subjectivity fully, the synchronism judge tends to bear sizable pressure.
Summary of the invention
In order to solve the stress problems of bringing that the artificial judgement method of prior art is given the synchronism referee; The present invention seeks to hope to alleviate synchronism judge's pressure through the method that adopts the computing machine automatic Evaluation; Improve the fairness of match; Realize a judgment system that can carry out objective appraisal to judge's level and fairness, for this reason, the present invention adopts the Computer Analysis technology to estimate the technical movements in the match; Utilize computer vision methods that synchronization of two-person synchronized diving is carried out automatic Evaluation Research, proposed an evaluation method that synchronism is analyzed automatically.
In order to realize described purpose, the object of the invention proposes a kind of method of automatic Evaluation synchronization of two-person synchronized diving, comprising:
Step 1: moving target extracts and is: rebuild based on dynamic background, from the synchronised diving action video, extract sportsman's outline and global camera motion parameter;
Step 2: synchronism characteristics is extracted and is: represent sportsman's synchronism characteristics JFF based on synchronism scoring key element in the diving rule, through sportsman's outline and the bottom-up calculating synchronism characteristics of the global camera motion parameter JFF value of having extracted;
Step 3: the synchronism evaluation is: synchronism characteristics JFF value is inserted the synchronism evaluation function, converts synchronism evaluation function value into, representes that according to the numerical values recited that obtains the height of synchronism is an evaluation result.
According to embodiments of the invention, said moving target extracts, and performing step comprises:
Step 11: to the dive video; Pass through overall motion estimation; Obtain the global camera motion parameter, output global motion parameter is used for synchronism characteristics and extracts; And utilize global motion compensation, try to achieve in the action video in each width of cloth picture frame sportsman's rough segmentation through adjacent image frame difference in the video and cut outline as output;
Step 12: utilize sportsman's rough segmentation to cut outline; Obtain the background redundant information in a plurality of continuous adjacent picture frames of each width of cloth picture frame in the action video, utilize the background redundant information in a plurality of continuous adjacent picture frames of each width of cloth picture frame to construct its color background image as output;
Step 13: the difference diagram of each width of cloth picture frame and its color background image in the calculating action video, difference diagram is carried out the test of hypothesis statistics obtain foreground area, the foreground area outline is that sportsman's outline is as output.
According to embodiments of the invention, said synchronism characteristics JFF adopts three JFF characteristics, is respectively that consistance characteristic JFF1, the action in the air of commencing height cooperates the consistance characteristic JFF2 of time and the consistance characteristic JFF3 of entry time.
According to embodiments of the invention, the bottom-up calculation procedure of said synchronism characteristics JFF value is following:
Step 21: is that the middle level characteristic is as output with low-level image feature through Feature Conversion; Low-level image feature is sportsman's outline and the global camera motion parameter of extracting; The middle level is characterized as vertical height, sportsman's outline variation characteristic vector sum entry mistiming of the relative first frame barycenter of sportsman's outline barycenter in every frame; Feature Conversion comprises that profile centroid calculation, specific region sportsman's outline surround the area proportional meter and calculate and the global camera motion compensation technique;
Step 22: the middle level characteristic is changed through the mark evaluation function, the mark evaluation function value that calculates is promptly exported synchronism characteristics JFF value as the high-level characteristic value.
According to embodiments of the invention; The consistance characteristic JFF1 value calculation procedure of the commencing height of said high-level characteristic JFF is following: according to sportsman's outline and global camera motion parameter low-level image feature; At first; Sportsman's outline utilization profile centroid calculation in each frame is drawn sportsman's outline barycenter; And through global camera motion compensation, sportsman's outline barycenter is projected to first frame, the vertical height that draws the relative first frame sportsman outline barycenter of sportsman's outline barycenter in every frame of on first frame, aliging of two sportsman's outline centroid positions in every frame is made the middle level characteristic; Insert corresponding mark evaluation function to vertical height and obtain high-level characteristic JFF1 value.
According to embodiments of the invention; The action in the air of said high-level characteristic JFF cooperates the consistance characteristic JFF2 value calculation procedure of time following: calculate according to sportsman's outline low-level image feature; At first, calculate specific region sportsman's area ratio that outline surrounds, set up polar coordinate system at sportsman's outline barycenter with the specific region pixels statistics; Pole axis with being spaced apart 90 degree is divided into four zones with sportsman's outline, and calculating sportsman's outline variation characteristic vector is that the middle level is characterized as:
F＝{fi}＝{f1，f2，f3，f4}，
Fi=Ni/Nsum wherein, Nsum is the number of all picture elements of sportsman's outline interior zone, and Ni is by the number of separated i intra-zone picture element, and what fi represented is that specific region sportsman's area ratio that outline surrounds is made the middle level characteristic; Insert corresponding mark evaluation function to sportsman's outline variation characteristic vector and obtain high-level characteristic JFF2 value.
According to embodiments of the invention; The consistance JFF3 value calculation procedure of the entry time of said high-level characteristic JFF is following: calculate according to low-level image features such as sportsman's outline and global camera motion parameters; At first; The vertical height similar with the process of the consistance characteristic JFF1 that asks commencing height, that the method for utilization profile centroid calculation and global motion compensation obtain the relative first frame sportsman outline barycenter of sportsman's outline barycenter in every frame; Then, when asking two sportsman's entry, relative position height between the two representes that the entry mistiming makes the middle level characteristic; Insert corresponding mark evaluation function to the entry mistiming and obtain high-level characteristic JFF3 value.
According to embodiments of the invention, the function body constitution step of said synchronism evaluation function comprises:
Step 31: according to synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations; It is belonged to same the action video in the match; The synchronous mark size of the evaluation of the judge between the combination in twos compares, and the synchronous mark between being made up in twos is order conduct output relatively;
Step 32: synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations is all carried out moving target extract with synchronism characteristics and extract, the synchronism characteristics JFF value conduct that obtains each action video is exported;
Belong to the synchronous mark of same the action video in the match between making up the in twos synchronism characteristics JFF value of order and each action video relatively in synchronised diving competition's action video of step 33 according to the synchronous mark of a plurality of known judges evaluations, the structure and parameter that employing preference property statistical learning method obtains synchronism evaluation function body is as output.
According to embodiments of the invention; Said step 31; In synchronised diving competition's action video of the synchronous mark that a plurality of known judges evaluate, being provided with A and B is same two action video in the match, if the synchronous score value A ' of the evaluation of the judge among the A is higher than the synchronous score value B ' of the judge's evaluation among the B; A '＞B ' is arranged, with the prior imformation of A '＞B ' as feedback element in the ranking functions body structure process.
Good effect of the present invention or advantage:
The present invention adopts the Computer Analysis technology to estimate the technical movements in the match; Utilize the method for computer vision that synchronization of two-person synchronized diving is carried out automatic Evaluation Research, experiment shows that it is very effective that this method is evaluated for the synchronization of two-person synchronized diving computing machine automatically.
The present invention is directed to synchronised diving evaluates automatically; Adopted a kind of can effective expression the synchronism characteristics JFF of synchronism scoring key element in the diving rule; Its advantage is: the JFF that from video, extracts through computing machine can be consistent with the factor that judge in the match considers, JFF comprises the reflection synchronism that one group of numerical value is can be from several aspects quantitative.Only consider in the function body construction process of synchronism evaluation function among the present invention in synchronised diving competition's action video of the synchronous mark that a plurality of known judges evaluate; The relative order of the synchronism mark of judge's evaluation between the action video in same match; Its advantage is: in the match of different plays, different referees has different opinion scales according to athletic integral level.The action of same quality possibly obtain different mark in different matches.Yet the final ranking of same match still can say something, and the mark possibility integral body of bout is all higher or whole on the low side, and still, the height of rank still can be represented the height of player motion level in the match of same field.The information of rank has been considered owing to have only in same the match in twos relatively, and the dives of different matches can be put in the training data goes.The analysis of also can putting together of synchronism between the dives of different matches.
Description of drawings
Fig. 1 is overall framework figure of the present invention;
Fig. 2 is that moving target extracts among the present invention;
Fig. 3 is that synchronism characteristics is extracted among the present invention;
Fig. 4 is synchronism evaluation among the present invention;
Fig. 5 is a synchronism evaluation function body structure process among the present invention;
Fig. 6 is a target leaching process result's of the present invention synoptic diagram;
Fig. 7 is the synoptic diagram that seethes character representation among the present invention;
Fig. 8 seethes changing features trend curve figure among the present invention;
Fig. 9 is the synchronised diving action inconsistency distribution schematic diagram of experimental result among the present invention.
Embodiment
Specify each related detailed problem in the technical scheme of the present invention below in conjunction with accompanying drawing.Be to be noted that described embodiment only is intended to be convenient to understanding of the present invention, and it is not played any qualification effect.
The algorithm use PC both can be realized among the present invention, suggestion computer configuration: processor: Pentium 4 CPU 3.00GHz, internal memory: 512MB.
Shown in Figure 1 like overall framework of the present invention: algorithm of the present invention comprises 3 parts, and (1) moving target extracts; (2) synchronism characteristics is extracted; (3) synchronism evaluation.
Extract shown in Figure 2 like moving target among the present invention: moving target extracts and comprises three step: the rough segmentation of step 11-image is cut, step 12-background is rebuild and step 13-foreground detection.Cut in the process in the rough segmentation of step 11 image; At first to the dive video; Draw the global camera motion parameter through overall motion estimation; Then on the basis of motion compensation to video in the adjacent image frame carry out difference and binaryzation, on binary map, carry out morphology at last and handle and to draw the player image rough segmentation and cut outline.The global motion parameter that process steps 11 obtains is cut in rough segmentation, is used for synchronism characteristics and extracts.It is to utilize redundant background information in a plurality of continuous adjacent picture frames of each width of cloth picture frame in the action video to construct the color background image of each width of cloth picture frame that step 12 background is rebuild.It is exactly that background information is cut in rough segmentation that the pixel that the profile representative is cut in sportsman's rough segmentation in the image is removed; Cut the global motion parameter that process steps 11 obtains according to rough segmentation again; Background information is cut in the rough segmentation of the adjacent several frames of each width of cloth picture frame project to present frame and obtain redundant background information, the mean value of getting consecutive frame projected background information at last is as current background information.In step 13 foreground detection, draw, confirm that with the way of test of hypothesis foreground area, its outline are sportsman's outline again through the differential chart that calculates each width of cloth picture frame and its color background image.That Fig. 6 target leaching process of the present invention result's synoptic diagram is represented is the result of the target leaching process of a certain frame in the action 307C action video; Sportsman's profile extracts; The upper left corner is original graph; The upper right corner is that the result is cut in the image rough segmentation, and the lower left corner is the result that background is rebuild, and the lower right corner is the result of foreground detection.
JFF (judging factor features) is a synchronism property characteristic.According to the diving rule that international swimming union announces, several key elements of synchronism marking comprise:
(1) take-off comprises the consistance of commencing height;
(2) action in the air cooperates the consistance of time;
(3) consistance of entry time.JFF is exactly the characteristic that can express above-mentioned factor.
Extract shown in Figure 3 like synchronism characteristics among the present invention: the present invention extracts JFF through a kind of bottom-up way from the low-level image feature to the high-level characteristic; Step 21: according to low-level image feature is that target is extracted sportsman's outline and the global camera motion parameter that obtains; Through calculating the area ratio that sportsman's outline is surrounded in target barycenter and the specific region; And the global motion compensation that carries out video camera carries out Feature Conversion, and the middle level characteristic that obtains is vertical height, sportsman's outline variation characteristic vector sum entry mistiming of the relative first frame barycenter of sportsman's outline barycenter in every frame; Step 22: the middle level characteristic is carried out the mark evaluation function calculate, convert the middle level eigenwert into synchronism characteristics JFF value that the high-level characteristic value is output; The present invention represents the synchronism factor of evaluation with 3 JFF, and synchronism characteristics is represented with factor of evaluation proper vector JFF=[JFF1, JFF2, JFF3].
Consider the consistance characteristic JFF1 of first factor commencing height; At first; The method of utilization profile centroid calculation calculates sportsman's outline barycenter; And method through global motion compensation; Sportsman's outline barycenter is projected to the first static frame of video camera, and two sportsman's outline centroid positions in every frame are alignd on first frame draws the vertical height position of relative first frame sportsman outline each time of barycenter of sportsman's outline barycenter in every frame, and obtaining movement objective orbit is the middle level characteristic; Insert corresponding mark evaluation function to vertical height and obtain high-level characteristic value JFF1.Adopt following evaluation function: JFF1=Δ height/bheight.Δ height is the absolute values of two sportsman's outline barycenter in the vertical drop of peak, and unit is a pixel.Bheght is a constant that converts pixel unit into standard unit (rice), in different videos, adopts different numerical as the case may be.
For second factor; Calculate through sportsman's outline low-level image feature; Calculate specific region sportsman's area ratio that outline surrounds with specific region pixels statistics method, the present invention has used a kind of special feature representation mode to estimate the consistance JFF2 that action in the air cooperates the time.At first, calculate specific region sportsman's area ratio that outline surrounds, set up polar coordinate system at athletic outline barycenter with the specific region pixels statistics.Pole axis with being spaced apart 90 degree is divided into four zones with profile, shown in the synoptic diagram that seethes character representation among Fig. 7 the present invention.
Calculating sportsman's outline variation characteristic vector is that the middle level is characterized as:
F＝{fi}＝{f1，f2，f3，f4}，
Fi=Ni/Nsum wherein, i=1,2,3,4; Nsum is the number of all picture elements of sportsman's outline interior zone, and Ni is by the number of separated i intra-zone picture element.What fi represented is that specific region sportsman's area ratio that outline surrounds is made the middle level characteristic; Obtain high-level characteristic value JFF2 to the corresponding mark evaluation function of sportsman's outline variation characteristic vector substitution.With variation tendency fs:fs=f1+f3 f1 and f3 and that represent sportsman's outline.When the sportsman accomplished a half cycle, the value of fs just reached a local peaking.What seethe among Fig. 8 the present invention that changing features trend curve figure representes is the variation tendency of seething when two sportsmen do certain 307C action in the same video, and different colours is represented different motion person.After the sportsman peaks, also just mean to begin to do the half cycle action, mark the position of the frame that reaches local peaking on the fs curve.Obtain after these positions, calculate evaluation function:
Wherein, hss is dive in requisition for the half cycle number that seethes (indicating to seethe 7 half cycles like 307C), and pframe (i) is fs picture frame ordinal position in action video in action video when reaching i peak value.
σ
k＝(pframe1(hss)-pframe1(1)+pframe2(hss)-pframe2(1))/(hss×2×2)
For the 3rd factor: the consistance JFF3 of entry time calculates according to low-level image features such as sportsman's outline and global camera motion parameters, and when estimating sportsman's entry, relative position height is between the two confirmed.Relative position height when confirming entry through two athletic height change curves.At first, similar with the process of the consistance characteristic JFF1 that asks commencing height, the method for utilization sportsman outline centroid calculation and the method for global motion compensation obtain the vertical height track of the outer relative first frame sportsman outline barycenter of barycenter of sportsman in every frame; Then, when asking sportsman's entry, relative position height between the two confirms that the entry mistiming makes the middle level characteristic; Obtain high-level characteristic value JFF3 to the corresponding mark evaluation function of entry mistiming substitution.Evaluation function: JFF3=Δ rheight/bheight.Wherein, Δ rheight is that two sportsman's outline barycenter relative height when entry is poor, and bheght is a constant that converts pixel unit into standard unit (rice), in different videos, adopts different numerical as the case may be.
Finally, shown in Figure 5 like evaluation function body structure process among synchronism evaluation procedure Fig. 4 and the present invention among the present invention: the present invention at first need construct the structure and parameter of synchronism evaluation function body according to synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations.Through calculating evaluation function, convert the synchronism characteristics value of action video into the evaluation function value, confirm the synchronism evaluation order according to the numerical values recited that obtains.
Step 31: according to synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations; It is belonged to same the action video in the match; The synchronous mark size of the evaluation of the judge between the combination in twos compares, and the synchronous mark between being made up in twos is order conduct output relatively;
Step 32: equal execution in step 1-moving target extraction (is rebuild based on dynamic background to synchronised diving competition's action video of the synchronous mark of a plurality of known judge's evaluations in the step 31; From the synchronised diving action video, extract sportsman's outline and global camera motion parameter) extract with step 2-synchronism characteristics and (represent sportsman's synchronism characteristics JFF based on the key element of marking of synchronism in the rule of diving; Through sportsman's outline and the bottom-up calculating synchronism characteristics of the global camera motion parameter JFF value of having extracted), the synchronism characteristics JFF value that obtains each action video is as output.
Step 33: according to step 31; Through synchronised diving competition's action video to the synchronous mark of a plurality of known judges evaluations; Its synchronous mark size that belongs to the judge's evaluation between making up in twos of action video in same the match compares; Synchronous mark between being made up in twos is order result and step 32 relatively; Synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations is all carried out moving target to be extracted and the synchronism characteristics extraction; Obtain the result of the synchronism characteristics JFF value of each action video, adopt the statistical learning method of preference inquiry learning strategy utilization RankBoost then, the structure and parameter that obtains the ranking functions body in the RankBoost method is:
Wherein, h
t(JFF) the single Weak Classifier function of expression, α
tRepresent the weight coefficient that single Weak Classifier function is corresponding, T representes the number of Weak Classifier function.h
t(JFF) and α
tArrive through the acquistion of RankBoost methodology.The H function is exactly the synchronism evaluation function.
The structure of evaluation function adopts the statistical learning method of preference inquiry learning strategy utilization RankBoost to obtain with synchronised diving competition's action video of the synchronous mark of a plurality of known judge's evaluations; Regard the evaluation of match as the problem of a preference inquiry learning; In the training process of evaluation function; A plurality of action video are made up in twos, only consider in synchronised diving competition's action video of the synchronous mark that a plurality of known judges evaluate, the relative order height of synchronism mark between the action video that in same play match, occurs makes up in twos; And the absolute phase difference mark between not considering to move in twos, and the action video between different competition sessions relatively.Consider the situation that everything makes up in twos in the known results match.Suppose in synchronised diving competition's action video of the synchronous mark that a plurality of known judges evaluate that being provided with action A is two action video in competing in same field with action B, the synchronous mark A ' of action A; Be 9 minutes, the synchronous mark B ' of action B is 8 minutes; In the ranking functions training process, the synchronism evaluation order of A of only will move promptly moves the synchronous mark A ' of A in the front of action B, greater than moving the synchronous mark B ' of B; With synchronous mark A '; As a prior imformation of feedback element in the ranking functions parameter learning process, train function parameters as a feedback information, for example move the synchronous mark A ' of A; Become 9.5, the synchronism evaluation order can't be affected for action A and action B.(action A synchronous mark A ', still greater than the action B synchronous mark B '.) again for example, the synchronous mark A ' of action A becomes 8.5 fens; The synchronous mark B ' of action B is 8 minutes, and the absolute phase difference mark between action A, action B has diminished; But; The synchronous mark A ' of action A still is higher than action B, and the prior imformation of A, action B generation of moving in the feedback element so can not be affected.
The present invention adopts the statistical learning method of preference inquiry learning strategy utilization RankBoost to obtain the structure and the parameter of evaluation function through synchronised diving competition's action video of the synchronous mark of a plurality of known judge's evaluations.In form, some data have been produced earlier to (v, f), wherein v is the dive video clips of known rank, and f is a JFF value of this section video.Definition sequencing feature ri is confirmed by corresponding JFF value ei; With evaluation of estimate ei ordering, suppose that ei j position, then has: ri ((v, f))=j.Definition deduction mark Score is the value of mean value gained of three effective marks of the synchronous mark of full marks known judge's evaluation of deducting the represented action of synchronised diving competition's action video segment in 10 minutes.Be higher than at the deduction mark Score* of v* under the deduction mark Score situation of v, the definition feedback function is: Φ v ((v*, f*), (v, f))=-1, Φ v ((v, f), (v*, f*))=+ 1, other situation, Φ v=0.So just can use the preferential learning method to draw the parameter of ranking functions H.The structure of ranking functions H and parameter declaration are at document Yoav Freund; Raj Iyer, RobertE.Schapire, Yoram Singer. An Efficient Boosting Algorithm forCombining Preference. Journal of Machine Learning Research; 2003 (4): introduce among the 933-969; The ranking functions H that obtains just as the synchronism evaluation function, is calculated the JFF characteristic through the dive video to unknown rank, calculate the value of corresponding evaluation function H then; The numerical values recited that obtains representes that the height of synchronism is an evaluation result, can carry out the synchronism rank to it according to evaluation result.
Implementation result
For the validity of verification method, designed one group of generation synchronization of data property sorting experiment of repeatedly sampling here.Synchronised diving competition's video of 30 2007 World Swimming Championships is adopted in experiment.Stochastic sampling is 20 synchronised diving competition's action video that are training sample as the synchronous mark of known judge's evaluation wherein, and remaining 10 is test sample book.Because the experimental data set that possibly generate altogether is a permutation and combination number, 10000 groups of unduplicated data of picked at random are as experimental data.At first, from the positive angle action video of diving tower, extract sportsman's outline.Then, obtain synchronism and judge needed synchronism characteristics JFF.At last, go out an evaluation function, come the synchronism between different dives in the evaluation test sample with evaluation function again with the method construct of training sample through RankBoost.
Adopt the error metrics method of inconsistency.Inconsistency disagreement is defined as:
Wherein, c representes a real number value function, is meant the synchronous mark of known judge's evaluation here.N representes all logarithms that can make up in the synchronism rank.[[... ]] symbolic representation: the result is 1 when the condition of centre is set up, otherwise the result is 0.X0, x1 are illustrated in same the match, the synchronism characteristics vector of any two groups of different diving videos.H representes the synchronism evaluation function.From Fig. 9 the present invention, can see that the distributional class of inconsistency is similar to normal distribution in the implementation result synchronised diving action inconsistency distribution schematic diagram.Wherein, average is 0.1083, and standard deviation is 0.0441.Can see that from table 1 the inconsistency error that surpasses the test data of half belongs to [0.075,0.125] interval.And the fiducial interval of [0.025,0.175] has comprised the test data more than 90%.The fiducial interval that can see the algorithm statistics from these data is smaller, and inconsistency is more concentrated, and algorithm can on average reach the accuracy about 90%.
From test data set, choosing more representational two groups analyzes.That table 2 is represented is the result of the 23rd group of data in the experiment, and that table 3 is represented is the result of the 189th group of data.Sequence number item in the table representes to be estimated the position of action video in these group data, and scoring item representes to be estimated the evaluation result of action video through evaluation function H.The deduction mark of the synchronous branch of the known judge's evaluation of a deduction of points expression.Therefore, the ordering here is actually the inverted order of true rank.Deduction of points number in the table 2 concentrates between 2.17 and 1.17.Here can see that the evaluation result between the identical deduction of points is not all in full accord, but the evaluation result between the different action of true score still can be good at ordering and makes a distinction.Deduction of points number in the table 3 then disperses relatively.Can find out that from this table true score differs the differentiation of sorting that evaluation result between bigger action can be apparent in view.
Table 1. inconsistency data
| Inconsistency is interval | How many intervals comprises data | Percent (100%) |
| 0～0.025 | ?149 | ?1.5 |
| 0.025～0.075 | ?1966 | ?19.7 |
| 0.075～0.125 | ?5035 | ?50.4 |
| 0.125～0.175 | ?2067 | ?20.7 |
| 0.175～0.225 | ?508 | ?5.1 |
| 0.225～0.275 | ?275 | ?2.8 |
Table 2. evaluation and score (1)
| Sequence number | Estimate | Deduction of points |
| 1 | 23.23703 | 2.83000 |
| 2 | 20.76646 | 2.67000 |
| 3 | 13.93426 | 2.17000 |
| 4 | 07.96665 | 2.17000 |
| 5 | 12.19469 | 2.17000 |
| 6 | 10.29751 | 2.17000 |
| 7 | 12.19469 | 2.00000 |
| 8 | 08.47920 | 1.17000 |
| 9 | 06.58413 | 1.17000 |
| 10 | 07.13067 | 1.17000 |
Table 3. evaluation and score (2)
| Sequence number | Estimate | Deduction of points |
| ?1 | ?18.06973 | ?2.67000 |
| ?2 | ?18.14224 | ?2.67000 |
| ?3 | ?15.99845 | ?2.50000 |
| ?4 | ?10.14118 | ?2.17000 |
| ?5 | ?09.92700 | ?2.17000 |
| ?6 | ?09.09735 | ?2.00000 |
| ?7 | ?09.92700 | ?2.00000 |
| ?8 | ?04.73707 | ?1.17000 |
| ?9 | ?02.18520 | ?1.00000 |
| ?10 | ?01.97984 | ?1.00000 |
The above; Be merely the embodiment among the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with this technological people in the technical scope that the present invention disclosed; Can understand conversion or the replacement expected; All should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (3)
1. the method for an automatic Evaluation synchronization of two-person synchronized diving is characterized in that, comprising:
Step 1: moving target extracts and is: rebuild based on dynamic background, from the synchronised diving action video, extract sportsman's outline and global camera motion parameter;
Step 2: synchronism characteristics is extracted and is: represent sportsman's synchronism characteristics JFF based on synchronism scoring key element in the diving rule, through sportsman's outline and the bottom-up calculating synchronism characteristics of the global camera motion parameter JFF value of having extracted;
Step 3: the synchronism evaluation is: synchronism characteristics JFF value is inserted the synchronism evaluation function, converts synchronism evaluation function value into, representes that according to the numerical values recited that obtains the height of synchronism is an evaluation result;
Said synchronism characteristics JFF adopts three JFF characteristics, is respectively that consistance characteristic JFF1, the action in the air of commencing height cooperates the consistance characteristic JFF2 of time and the consistance characteristic JFF3 of entry time;
The bottom-up calculation procedure of said synchronism characteristics JFF value is following:
Step 21: is that the middle level characteristic is as output with low-level image feature through Feature Conversion; Low-level image feature is sportsman's outline and the global camera motion parameter of extracting; The middle level is characterized as vertical height, sportsman's outline variation characteristic vector sum entry mistiming of the relative first frame barycenter of sportsman's outline barycenter in every frame; Feature Conversion comprises that profile centroid calculation, specific region sportsman's outline surround the area proportional meter and calculate and the global camera motion compensation technique;
Step 22: the middle level characteristic is changed through the mark evaluation function, the mark evaluation function value that calculates is promptly exported synchronism characteristics JFF value as the high-level characteristic value;
The consistance characteristic JFF1 value calculation procedure of the commencing height of said synchronism characteristics JFF is following: according to sportsman's outline and global camera motion parameter low-level image feature; At first; Sportsman's outline utilization profile centroid calculation in each frame is drawn sportsman's outline barycenter; And through the global camera motion compensation; Sportsman's outline barycenter is projected to first frame, on first frame, the align vertical height that draws the relative first frame sportsman outline barycenter of sportsman's outline barycenter in every frame of two sportsman's outline centroid positions in every frame is made the middle level characteristic and inserted corresponding mark evaluation function to vertical height and obtain high-level characteristic JFF1 value;
The action in the air of said synchronism characteristics JFF cooperates the consistance characteristic JFF2 value calculation procedure of time following: calculate according to sportsman's outline low-level image feature; At first; Calculate specific region sportsman's area ratio that outline surrounds with the specific region pixels statistics; Set up polar coordinate system at sportsman's outline barycenter, with the pole axis that is spaced apart 90 degree sportsman's outline is divided into four zones, calculating sportsman's outline variation characteristic vector is that the middle level is characterized as:
F＝{fi}＝{f1，f2，f3，f4}，
Fi=Ni/Nsum wherein; Nsum is the number of all picture elements of sportsman's outline interior zone; Ni is by the number of separated i intra-zone picture element, and fi representes is that specific region sportsman's area ratio that outline surrounds is made the middle level characteristic and inserted corresponding mark evaluation function to sportsman's outline variation characteristic vector and obtain high-level characteristic JFF2 value;
The consistance JFF3 value calculation procedure of the entry time of said synchronism characteristics JFF is following: calculate according to low-level image features such as sportsman's outline and global camera motion parameters; At first; The vertical height similar with the process of the consistance characteristic JFF1 that asks commencing height, that the method for utilization profile centroid calculation and global motion compensation obtain the relative first frame sportsman outline barycenter of sportsman's outline barycenter in every frame; Then, when asking two sportsman's entry, relative position height between the two representes that the entry mistiming makes the middle level characteristic and insert corresponding mark evaluation function to the entry mistiming and obtain high-level characteristic JFF3 value;
The function body constitution step of said synchronism evaluation function comprises:
Step 31: according to synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations; It is belonged to same the action video in the match; The synchronous mark size of the evaluation of the judge between the combination in twos compares, and the synchronous mark between being made up in twos is order conduct output relatively;
Step 32: synchronised diving competition's action video of the synchronous mark of a plurality of known judges evaluations is all carried out moving target extract with synchronism characteristics and extract, the synchronism characteristics JFF value conduct that obtains each action video is exported;
Step 33: belong to the synchronous mark of same the action video in the match between making up the in twos synchronism characteristics JFF value of order and each action video relatively in the synchronised diving competition's action video according to the synchronous mark of a plurality of known judges evaluations, the structure and parameter that employing preference property statistical learning method obtains synchronism evaluation function body is as output.
2. the method for automatic Evaluation synchronization of two-person synchronized diving according to claim 1 is characterized in that, said moving target extracts, and performing step comprises:
Step 11: to the dive video; Pass through overall motion estimation; Obtain the global camera motion parameter, output global motion parameter is used for synchronism characteristics and extracts; And utilize global motion compensation, try to achieve in the action video in each width of cloth picture frame sportsman's rough segmentation through adjacent image frame difference in the video and cut outline as output;
Step 12: utilize sportsman's rough segmentation to cut outline; Obtain the background redundant information in a plurality of continuous adjacent picture frames of each width of cloth picture frame in the action video, utilize the background redundant information in a plurality of continuous adjacent picture frames of each width of cloth picture frame to construct its color background image as output;
Step 13: the difference diagram of each width of cloth picture frame and its color background image in the calculating action video, difference diagram is carried out the test of hypothesis statistics obtain foreground area, the foreground area outline is that sportsman's outline is as output.
3. the method for automatic Evaluation synchronization of two-person synchronized diving according to claim 1; It is characterized in that said step 31 is in synchronised diving competition's action video of the synchronous mark that a plurality of known judges evaluate; Being provided with A and B is same two action video in the match; If the synchronous score value A ' of the evaluation of the judge among the A is higher than the synchronous score value B ' of the judge's evaluation among the B, A '＞B ' is arranged, with the prior imformation of A '＞B ' as feedback element in the ranking functions body structure process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2007103042260A CN101470898B (en) | 2007-12-26 | 2007-12-26 | Automatic analysis method for synchronization of two-person synchronized diving |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2007103042260A CN101470898B (en) | 2007-12-26 | 2007-12-26 | Automatic analysis method for synchronization of two-person synchronized diving |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101470898A CN101470898A (en) | 2009-07-01 |
| CN101470898B true CN101470898B (en) | 2012-04-11 |
Family
ID=40828351
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2007103042260A Expired - Fee Related CN101470898B (en) | 2007-12-26 | 2007-12-26 | Automatic analysis method for synchronization of two-person synchronized diving |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101470898B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102281385B (en) * | 2011-08-16 | 2013-03-27 | 上海交通大学 | Periodic motion detection method based on motion video |
| RU2492898C2 (en) * | 2011-09-21 | 2013-09-20 | Общество с ограниченной ответственностью "Фигурное катание. Наука побеждать" | Method of measuring parameters of rotational and translational motions of figure-skaters in pair skating |
| EP2982422A1 (en) * | 2013-04-02 | 2016-02-10 | NEC Solution Innovators, Ltd. | Body-motion assessment device, dance assessment device, karaoke device, and game device |
| GB201703243D0 (en) * | 2017-02-28 | 2017-04-12 | Pro Sport Support Ltd | System, method, apparatus and marker for assessing athletic performance |
| WO2018211571A1 (en) * | 2017-05-15 | 2018-11-22 | 富士通株式会社 | Performance display program, performance display method, and performance display device |
| CN108744414B (en) * | 2018-04-09 | 2020-01-03 | 广东斐瑞智能技术有限公司 | Exercise physical therapy apparatus and exercise evaluation method |
| CN108905095B (en) * | 2018-08-06 | 2020-03-10 | 百度在线网络技术（北京）有限公司 | Athlete competition state evaluation method and equipment |
| CN110222977A (en) * | 2019-06-03 | 2019-09-10 | 张学志 | One kind movement sport methods of marking based on computer vision and device |
| CN110327611A (en) * | 2019-06-17 | 2019-10-15 | 杭州电子科技大学 | A kind of platform diving evaluation method that athletic type's coefficient is added |
| CN113033968B (en) * | 2021-03-04 | 2022-03-25 | 深圳旗鱼体育传播有限公司 | Method for evaluating competition performance of competition players in sailing boat competition |
| CN113343774B (en) * | 2021-05-13 | 2022-11-01 | 东南大学 | Fine-grained engineering mechanics diving action simulation and evaluation method |
| CN114679455B (en) * | 2022-03-27 | 2022-11-08 | 江苏海纳宝川智能科技有限公司 | Distributed cloud service system |
| CN114722230B (en) * | 2022-03-30 | 2022-10-18 | 西安西科软件技术有限公司 | Auxiliary judgment system using angle big data matching |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5745029A (en) * | 1992-09-03 | 1998-04-28 | Manuel Garcia | Refereeing aid system for ball games |
| GB2339504A (en) * | 1998-04-25 | 2000-01-26 | Peter Russell Murray | Referee's computer aid |
| CN1150477C (en) * | 2001-04-18 | 2004-05-19 | 中国科学院计算技术研究所 | Method for computer processing, live broadcast-over-network and auxiliary judgement of in-situ match information |
| CN2774590Y (en) * | 2005-03-30 | 2006-04-26 | 中国海洋大学 | Cross border alarm for tennis game |
-
2007
- 2007-12-26 CN CN2007103042260A patent/CN101470898B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5745029A (en) * | 1992-09-03 | 1998-04-28 | Manuel Garcia | Refereeing aid system for ball games |
| GB2339504A (en) * | 1998-04-25 | 2000-01-26 | Peter Russell Murray | Referee's computer aid |
| CN1150477C (en) * | 2001-04-18 | 2004-05-19 | 中国科学院计算技术研究所 | Method for computer processing, live broadcast-over-network and auxiliary judgement of in-situ match information |
| CN2774590Y (en) * | 2005-03-30 | 2006-04-26 | 中国海洋大学 | Cross border alarm for tennis game |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101470898A (en) | 2009-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101470898B (en) | Automatic analysis method for synchronization of two-person synchronized diving | |
| CN109948459B (en) | Football action evaluation method and system based on deep learning | |
| Li et al. | Intelligent sports training system based on artificial intelligence and big data | |
| CN110298332A (en) | Method, system, computer equipment and the storage medium of Activity recognition | |
| CN112819852A (en) | Evaluating gesture-based motion | |
| Morais et al. | Cluster stability as a new method to assess changes in performance and its determinant factors over a season in young swimmers | |
| CN109045664B (en) | Diving scoring method, server and system based on deep learning | |
| CN107396095A (en) | One kind is without with reference to three-dimensional image quality evaluation method | |
| Rathi et al. | Applications of artificial intelligence in the game of football: The global perspective | |
| CN109248428A (en) | A kind of dynamic analysing method of tennis trajectory processing system | |
| CN110458100A (en) | Based on target detection and the identification of the table tennis drop point of tracking and methods of marking and system | |
| Young et al. | Judging complex movement performances for excellence: a principal components analysis-based technique applied to competitive diving | |
| CN111905350B (en) | Automatic table tennis hitting performance evaluation method and system based on motion data | |
| Hu et al. | Basketball activity classification based on upper body kinematics and dynamic time warping | |
| CN110490978A (en) | Outdoor scene based on mixed reality technology is ridden training method | |
| Hsu et al. | Coachai: A project for microscopic badminton match data collection and tactical analysis | |
| Yang et al. | Research on face recognition sports intelligence training platform based on artificial intelligence | |
| CN112933581A (en) | Sports action scoring method and device based on virtual reality technology | |
| Li et al. | 3D-Yoga: a 3D yoga dataset for visual-based hierarchical sports action analysis | |
| CN114550071B (en) | Method, device and medium for automatically identifying and capturing track and field video action key frames | |
| CN113975775B (en) | Wearable inertial body feeling ping-pong exercise training system and working method thereof | |
| CN115311745A (en) | Pattern skating layered action recognition method | |
| Moness et al. | Selecting Promising Junior Swimmers in Egypt Using Automated Biometric Algorithms of Image Processing and Fuzzy Concepts | |
| CN109977819A (en) | A kind of Weakly supervised individual part localization method of application template matching process | |
| Baca et al. | 13th World Congress of Performance Analysis of Sport and 13th International Symposium on Computer Science in Sport |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120411 Termination date: 20211226 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |